]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
9ea459e1 | 26 | #include <linux/task_work.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
d2921684 EB |
30 | /* Maximum number of mounts in a mount namespace */ |
31 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
32 | ||
0818bf27 AV |
33 | static unsigned int m_hash_mask __read_mostly; |
34 | static unsigned int m_hash_shift __read_mostly; | |
35 | static unsigned int mp_hash_mask __read_mostly; | |
36 | static unsigned int mp_hash_shift __read_mostly; | |
37 | ||
38 | static __initdata unsigned long mhash_entries; | |
39 | static int __init set_mhash_entries(char *str) | |
40 | { | |
41 | if (!str) | |
42 | return 0; | |
43 | mhash_entries = simple_strtoul(str, &str, 0); | |
44 | return 1; | |
45 | } | |
46 | __setup("mhash_entries=", set_mhash_entries); | |
47 | ||
48 | static __initdata unsigned long mphash_entries; | |
49 | static int __init set_mphash_entries(char *str) | |
50 | { | |
51 | if (!str) | |
52 | return 0; | |
53 | mphash_entries = simple_strtoul(str, &str, 0); | |
54 | return 1; | |
55 | } | |
56 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 57 | |
c7999c36 | 58 | static u64 event; |
73cd49ec | 59 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 60 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 61 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
62 | static int mnt_id_start = 0; |
63 | static int mnt_group_start = 1; | |
1da177e4 | 64 | |
38129a13 | 65 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 66 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 67 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 68 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 69 | |
f87fd4c2 | 70 | /* /sys/fs */ |
00d26666 GKH |
71 | struct kobject *fs_kobj; |
72 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 73 | |
99b7db7b NP |
74 | /* |
75 | * vfsmount lock may be taken for read to prevent changes to the | |
76 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
77 | * up the tree. | |
78 | * | |
79 | * It should be taken for write in all cases where the vfsmount | |
80 | * tree or hash is modified or when a vfsmount structure is modified. | |
81 | */ | |
48a066e7 | 82 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 83 | |
38129a13 | 84 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 85 | { |
b58fed8b RP |
86 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
87 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
88 | tmp = tmp + (tmp >> m_hash_shift); |
89 | return &mount_hashtable[tmp & m_hash_mask]; | |
90 | } | |
91 | ||
92 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
93 | { | |
94 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
95 | tmp = tmp + (tmp >> mp_hash_shift); | |
96 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
97 | } |
98 | ||
b105e270 | 99 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
100 | { |
101 | int res; | |
102 | ||
103 | retry: | |
104 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 105 | spin_lock(&mnt_id_lock); |
15169fe7 | 106 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 107 | if (!res) |
15169fe7 | 108 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 109 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
110 | if (res == -EAGAIN) |
111 | goto retry; | |
112 | ||
113 | return res; | |
114 | } | |
115 | ||
b105e270 | 116 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 117 | { |
15169fe7 | 118 | int id = mnt->mnt_id; |
99b7db7b | 119 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
120 | ida_remove(&mnt_id_ida, id); |
121 | if (mnt_id_start > id) | |
122 | mnt_id_start = id; | |
99b7db7b | 123 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
124 | } |
125 | ||
719f5d7f MS |
126 | /* |
127 | * Allocate a new peer group ID | |
128 | * | |
129 | * mnt_group_ida is protected by namespace_sem | |
130 | */ | |
4b8b21f4 | 131 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 132 | { |
f21f6220 AV |
133 | int res; |
134 | ||
719f5d7f MS |
135 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
136 | return -ENOMEM; | |
137 | ||
f21f6220 AV |
138 | res = ida_get_new_above(&mnt_group_ida, |
139 | mnt_group_start, | |
15169fe7 | 140 | &mnt->mnt_group_id); |
f21f6220 | 141 | if (!res) |
15169fe7 | 142 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
143 | |
144 | return res; | |
719f5d7f MS |
145 | } |
146 | ||
147 | /* | |
148 | * Release a peer group ID | |
149 | */ | |
4b8b21f4 | 150 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 151 | { |
15169fe7 | 152 | int id = mnt->mnt_group_id; |
f21f6220 AV |
153 | ida_remove(&mnt_group_ida, id); |
154 | if (mnt_group_start > id) | |
155 | mnt_group_start = id; | |
15169fe7 | 156 | mnt->mnt_group_id = 0; |
719f5d7f MS |
157 | } |
158 | ||
b3e19d92 NP |
159 | /* |
160 | * vfsmount lock must be held for read | |
161 | */ | |
83adc753 | 162 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
163 | { |
164 | #ifdef CONFIG_SMP | |
68e8a9fe | 165 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
166 | #else |
167 | preempt_disable(); | |
68e8a9fe | 168 | mnt->mnt_count += n; |
b3e19d92 NP |
169 | preempt_enable(); |
170 | #endif | |
171 | } | |
172 | ||
b3e19d92 NP |
173 | /* |
174 | * vfsmount lock must be held for write | |
175 | */ | |
83adc753 | 176 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
177 | { |
178 | #ifdef CONFIG_SMP | |
f03c6599 | 179 | unsigned int count = 0; |
b3e19d92 NP |
180 | int cpu; |
181 | ||
182 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 183 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
184 | } |
185 | ||
186 | return count; | |
187 | #else | |
68e8a9fe | 188 | return mnt->mnt_count; |
b3e19d92 NP |
189 | #endif |
190 | } | |
191 | ||
87b95ce0 AV |
192 | static void drop_mountpoint(struct fs_pin *p) |
193 | { | |
194 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
195 | dput(m->mnt_ex_mountpoint); | |
196 | pin_remove(p); | |
197 | mntput(&m->mnt); | |
198 | } | |
199 | ||
b105e270 | 200 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 201 | { |
c63181e6 AV |
202 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
203 | if (mnt) { | |
73cd49ec MS |
204 | int err; |
205 | ||
c63181e6 | 206 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
207 | if (err) |
208 | goto out_free_cache; | |
209 | ||
210 | if (name) { | |
fcc139ae | 211 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 212 | if (!mnt->mnt_devname) |
88b38782 | 213 | goto out_free_id; |
73cd49ec MS |
214 | } |
215 | ||
b3e19d92 | 216 | #ifdef CONFIG_SMP |
c63181e6 AV |
217 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
218 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
219 | goto out_free_devname; |
220 | ||
c63181e6 | 221 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 222 | #else |
c63181e6 AV |
223 | mnt->mnt_count = 1; |
224 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
225 | #endif |
226 | ||
38129a13 | 227 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
228 | INIT_LIST_HEAD(&mnt->mnt_child); |
229 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
230 | INIT_LIST_HEAD(&mnt->mnt_list); | |
231 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
232 | INIT_LIST_HEAD(&mnt->mnt_share); | |
233 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
234 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 235 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
236 | #ifdef CONFIG_FSNOTIFY |
237 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 238 | #endif |
87b95ce0 | 239 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 240 | } |
c63181e6 | 241 | return mnt; |
88b38782 | 242 | |
d3ef3d73 | 243 | #ifdef CONFIG_SMP |
244 | out_free_devname: | |
fcc139ae | 245 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 246 | #endif |
88b38782 | 247 | out_free_id: |
c63181e6 | 248 | mnt_free_id(mnt); |
88b38782 | 249 | out_free_cache: |
c63181e6 | 250 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 251 | return NULL; |
1da177e4 LT |
252 | } |
253 | ||
3d733633 DH |
254 | /* |
255 | * Most r/o checks on a fs are for operations that take | |
256 | * discrete amounts of time, like a write() or unlink(). | |
257 | * We must keep track of when those operations start | |
258 | * (for permission checks) and when they end, so that | |
259 | * we can determine when writes are able to occur to | |
260 | * a filesystem. | |
261 | */ | |
262 | /* | |
263 | * __mnt_is_readonly: check whether a mount is read-only | |
264 | * @mnt: the mount to check for its write status | |
265 | * | |
266 | * This shouldn't be used directly ouside of the VFS. | |
267 | * It does not guarantee that the filesystem will stay | |
268 | * r/w, just that it is right *now*. This can not and | |
269 | * should not be used in place of IS_RDONLY(inode). | |
270 | * mnt_want/drop_write() will _keep_ the filesystem | |
271 | * r/w. | |
272 | */ | |
273 | int __mnt_is_readonly(struct vfsmount *mnt) | |
274 | { | |
2e4b7fcd DH |
275 | if (mnt->mnt_flags & MNT_READONLY) |
276 | return 1; | |
277 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
278 | return 1; | |
279 | return 0; | |
3d733633 DH |
280 | } |
281 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
282 | ||
83adc753 | 283 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 284 | { |
285 | #ifdef CONFIG_SMP | |
68e8a9fe | 286 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 287 | #else |
68e8a9fe | 288 | mnt->mnt_writers++; |
d3ef3d73 | 289 | #endif |
290 | } | |
3d733633 | 291 | |
83adc753 | 292 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 293 | { |
d3ef3d73 | 294 | #ifdef CONFIG_SMP |
68e8a9fe | 295 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 296 | #else |
68e8a9fe | 297 | mnt->mnt_writers--; |
d3ef3d73 | 298 | #endif |
3d733633 | 299 | } |
3d733633 | 300 | |
83adc753 | 301 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 302 | { |
d3ef3d73 | 303 | #ifdef CONFIG_SMP |
304 | unsigned int count = 0; | |
3d733633 | 305 | int cpu; |
3d733633 DH |
306 | |
307 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 308 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 309 | } |
3d733633 | 310 | |
d3ef3d73 | 311 | return count; |
312 | #else | |
313 | return mnt->mnt_writers; | |
314 | #endif | |
3d733633 DH |
315 | } |
316 | ||
4ed5e82f MS |
317 | static int mnt_is_readonly(struct vfsmount *mnt) |
318 | { | |
319 | if (mnt->mnt_sb->s_readonly_remount) | |
320 | return 1; | |
321 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
322 | smp_rmb(); | |
323 | return __mnt_is_readonly(mnt); | |
324 | } | |
325 | ||
8366025e | 326 | /* |
eb04c282 JK |
327 | * Most r/o & frozen checks on a fs are for operations that take discrete |
328 | * amounts of time, like a write() or unlink(). We must keep track of when | |
329 | * those operations start (for permission checks) and when they end, so that we | |
330 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
331 | */ |
332 | /** | |
eb04c282 | 333 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 334 | * @m: the mount on which to take a write |
8366025e | 335 | * |
eb04c282 JK |
336 | * This tells the low-level filesystem that a write is about to be performed to |
337 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
338 | * returning success. This operation does not protect against filesystem being | |
339 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
340 | * called. This is effectively a refcount. | |
8366025e | 341 | */ |
eb04c282 | 342 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 343 | { |
83adc753 | 344 | struct mount *mnt = real_mount(m); |
3d733633 | 345 | int ret = 0; |
3d733633 | 346 | |
d3ef3d73 | 347 | preempt_disable(); |
c6653a83 | 348 | mnt_inc_writers(mnt); |
d3ef3d73 | 349 | /* |
c6653a83 | 350 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 351 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
352 | * incremented count after it has set MNT_WRITE_HOLD. | |
353 | */ | |
354 | smp_mb(); | |
1e75529e | 355 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 356 | cpu_relax(); |
357 | /* | |
358 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
359 | * be set to match its requirements. So we must not load that until | |
360 | * MNT_WRITE_HOLD is cleared. | |
361 | */ | |
362 | smp_rmb(); | |
4ed5e82f | 363 | if (mnt_is_readonly(m)) { |
c6653a83 | 364 | mnt_dec_writers(mnt); |
3d733633 | 365 | ret = -EROFS; |
3d733633 | 366 | } |
d3ef3d73 | 367 | preempt_enable(); |
eb04c282 JK |
368 | |
369 | return ret; | |
370 | } | |
371 | ||
372 | /** | |
373 | * mnt_want_write - get write access to a mount | |
374 | * @m: the mount on which to take a write | |
375 | * | |
376 | * This tells the low-level filesystem that a write is about to be performed to | |
377 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
378 | * is not frozen) before returning success. When the write operation is | |
379 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
380 | */ | |
381 | int mnt_want_write(struct vfsmount *m) | |
382 | { | |
383 | int ret; | |
384 | ||
385 | sb_start_write(m->mnt_sb); | |
386 | ret = __mnt_want_write(m); | |
387 | if (ret) | |
388 | sb_end_write(m->mnt_sb); | |
3d733633 | 389 | return ret; |
8366025e DH |
390 | } |
391 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
392 | ||
96029c4e | 393 | /** |
394 | * mnt_clone_write - get write access to a mount | |
395 | * @mnt: the mount on which to take a write | |
396 | * | |
397 | * This is effectively like mnt_want_write, except | |
398 | * it must only be used to take an extra write reference | |
399 | * on a mountpoint that we already know has a write reference | |
400 | * on it. This allows some optimisation. | |
401 | * | |
402 | * After finished, mnt_drop_write must be called as usual to | |
403 | * drop the reference. | |
404 | */ | |
405 | int mnt_clone_write(struct vfsmount *mnt) | |
406 | { | |
407 | /* superblock may be r/o */ | |
408 | if (__mnt_is_readonly(mnt)) | |
409 | return -EROFS; | |
410 | preempt_disable(); | |
83adc753 | 411 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 412 | preempt_enable(); |
413 | return 0; | |
414 | } | |
415 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
416 | ||
417 | /** | |
eb04c282 | 418 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 419 | * @file: the file who's mount on which to take a write |
420 | * | |
eb04c282 | 421 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 422 | * do some optimisations if the file is open for write already |
423 | */ | |
eb04c282 | 424 | int __mnt_want_write_file(struct file *file) |
96029c4e | 425 | { |
83f936c7 | 426 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 427 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 428 | else |
429 | return mnt_clone_write(file->f_path.mnt); | |
430 | } | |
eb04c282 JK |
431 | |
432 | /** | |
433 | * mnt_want_write_file - get write access to a file's mount | |
434 | * @file: the file who's mount on which to take a write | |
435 | * | |
436 | * This is like mnt_want_write, but it takes a file and can | |
437 | * do some optimisations if the file is open for write already | |
438 | */ | |
439 | int mnt_want_write_file(struct file *file) | |
440 | { | |
441 | int ret; | |
442 | ||
443 | sb_start_write(file->f_path.mnt->mnt_sb); | |
444 | ret = __mnt_want_write_file(file); | |
445 | if (ret) | |
446 | sb_end_write(file->f_path.mnt->mnt_sb); | |
447 | return ret; | |
448 | } | |
96029c4e | 449 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
450 | ||
8366025e | 451 | /** |
eb04c282 | 452 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
453 | * @mnt: the mount on which to give up write access |
454 | * | |
455 | * Tells the low-level filesystem that we are done | |
456 | * performing writes to it. Must be matched with | |
eb04c282 | 457 | * __mnt_want_write() call above. |
8366025e | 458 | */ |
eb04c282 | 459 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 460 | { |
d3ef3d73 | 461 | preempt_disable(); |
83adc753 | 462 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 463 | preempt_enable(); |
8366025e | 464 | } |
e14748e8 | 465 | EXPORT_SYMBOL_GPL(__mnt_drop_write); |
eb04c282 JK |
466 | |
467 | /** | |
468 | * mnt_drop_write - give up write access to a mount | |
469 | * @mnt: the mount on which to give up write access | |
470 | * | |
471 | * Tells the low-level filesystem that we are done performing writes to it and | |
472 | * also allows filesystem to be frozen again. Must be matched with | |
473 | * mnt_want_write() call above. | |
474 | */ | |
475 | void mnt_drop_write(struct vfsmount *mnt) | |
476 | { | |
477 | __mnt_drop_write(mnt); | |
478 | sb_end_write(mnt->mnt_sb); | |
479 | } | |
8366025e DH |
480 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
481 | ||
eb04c282 JK |
482 | void __mnt_drop_write_file(struct file *file) |
483 | { | |
484 | __mnt_drop_write(file->f_path.mnt); | |
485 | } | |
486 | ||
2a79f17e AV |
487 | void mnt_drop_write_file(struct file *file) |
488 | { | |
489 | mnt_drop_write(file->f_path.mnt); | |
490 | } | |
491 | EXPORT_SYMBOL(mnt_drop_write_file); | |
492 | ||
83adc753 | 493 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 494 | { |
3d733633 DH |
495 | int ret = 0; |
496 | ||
719ea2fb | 497 | lock_mount_hash(); |
83adc753 | 498 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 499 | /* |
d3ef3d73 | 500 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
501 | * should be visible before we do. | |
3d733633 | 502 | */ |
d3ef3d73 | 503 | smp_mb(); |
504 | ||
3d733633 | 505 | /* |
d3ef3d73 | 506 | * With writers on hold, if this value is zero, then there are |
507 | * definitely no active writers (although held writers may subsequently | |
508 | * increment the count, they'll have to wait, and decrement it after | |
509 | * seeing MNT_READONLY). | |
510 | * | |
511 | * It is OK to have counter incremented on one CPU and decremented on | |
512 | * another: the sum will add up correctly. The danger would be when we | |
513 | * sum up each counter, if we read a counter before it is incremented, | |
514 | * but then read another CPU's count which it has been subsequently | |
515 | * decremented from -- we would see more decrements than we should. | |
516 | * MNT_WRITE_HOLD protects against this scenario, because | |
517 | * mnt_want_write first increments count, then smp_mb, then spins on | |
518 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
519 | * we're counting up here. | |
3d733633 | 520 | */ |
c6653a83 | 521 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 522 | ret = -EBUSY; |
523 | else | |
83adc753 | 524 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 525 | /* |
526 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
527 | * that become unheld will see MNT_READONLY. | |
528 | */ | |
529 | smp_wmb(); | |
83adc753 | 530 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 531 | unlock_mount_hash(); |
3d733633 | 532 | return ret; |
8366025e | 533 | } |
8366025e | 534 | |
83adc753 | 535 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 536 | { |
719ea2fb | 537 | lock_mount_hash(); |
83adc753 | 538 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 539 | unlock_mount_hash(); |
2e4b7fcd DH |
540 | } |
541 | ||
4ed5e82f MS |
542 | int sb_prepare_remount_readonly(struct super_block *sb) |
543 | { | |
544 | struct mount *mnt; | |
545 | int err = 0; | |
546 | ||
8e8b8796 MS |
547 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
548 | if (atomic_long_read(&sb->s_remove_count)) | |
549 | return -EBUSY; | |
550 | ||
719ea2fb | 551 | lock_mount_hash(); |
4ed5e82f MS |
552 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
553 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
554 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
555 | smp_mb(); | |
556 | if (mnt_get_writers(mnt) > 0) { | |
557 | err = -EBUSY; | |
558 | break; | |
559 | } | |
560 | } | |
561 | } | |
8e8b8796 MS |
562 | if (!err && atomic_long_read(&sb->s_remove_count)) |
563 | err = -EBUSY; | |
564 | ||
4ed5e82f MS |
565 | if (!err) { |
566 | sb->s_readonly_remount = 1; | |
567 | smp_wmb(); | |
568 | } | |
569 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
570 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
571 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
572 | } | |
719ea2fb | 573 | unlock_mount_hash(); |
4ed5e82f MS |
574 | |
575 | return err; | |
576 | } | |
577 | ||
b105e270 | 578 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 579 | { |
fcc139ae | 580 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 581 | #ifdef CONFIG_SMP |
68e8a9fe | 582 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 583 | #endif |
b105e270 | 584 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
585 | } |
586 | ||
8ffcb32e DH |
587 | static void delayed_free_vfsmnt(struct rcu_head *head) |
588 | { | |
589 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
590 | } | |
591 | ||
48a066e7 | 592 | /* call under rcu_read_lock */ |
294d71ff | 593 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
594 | { |
595 | struct mount *mnt; | |
596 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 597 | return 1; |
48a066e7 | 598 | if (bastard == NULL) |
294d71ff | 599 | return 0; |
48a066e7 AV |
600 | mnt = real_mount(bastard); |
601 | mnt_add_count(mnt, 1); | |
602 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 603 | return 0; |
48a066e7 AV |
604 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
605 | mnt_add_count(mnt, -1); | |
294d71ff AV |
606 | return 1; |
607 | } | |
608 | return -1; | |
609 | } | |
610 | ||
611 | /* call under rcu_read_lock */ | |
612 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
613 | { | |
614 | int res = __legitimize_mnt(bastard, seq); | |
615 | if (likely(!res)) | |
616 | return true; | |
617 | if (unlikely(res < 0)) { | |
618 | rcu_read_unlock(); | |
619 | mntput(bastard); | |
620 | rcu_read_lock(); | |
48a066e7 | 621 | } |
48a066e7 AV |
622 | return false; |
623 | } | |
624 | ||
1da177e4 | 625 | /* |
474279dc | 626 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 627 | * call under rcu_read_lock() |
1da177e4 | 628 | */ |
474279dc | 629 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 630 | { |
38129a13 | 631 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
632 | struct mount *p; |
633 | ||
38129a13 | 634 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
635 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
636 | return p; | |
637 | return NULL; | |
638 | } | |
639 | ||
640 | /* | |
641 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 642 | * mount_lock must be held. |
474279dc AV |
643 | */ |
644 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
645 | { | |
411a938b EB |
646 | struct mount *p, *res = NULL; |
647 | p = __lookup_mnt(mnt, dentry); | |
38129a13 AV |
648 | if (!p) |
649 | goto out; | |
411a938b EB |
650 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
651 | res = p; | |
38129a13 | 652 | hlist_for_each_entry_continue(p, mnt_hash) { |
1d6a32ac AV |
653 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
654 | break; | |
411a938b EB |
655 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
656 | res = p; | |
1d6a32ac | 657 | } |
38129a13 | 658 | out: |
1d6a32ac | 659 | return res; |
1da177e4 LT |
660 | } |
661 | ||
a05964f3 | 662 | /* |
f015f126 DH |
663 | * lookup_mnt - Return the first child mount mounted at path |
664 | * | |
665 | * "First" means first mounted chronologically. If you create the | |
666 | * following mounts: | |
667 | * | |
668 | * mount /dev/sda1 /mnt | |
669 | * mount /dev/sda2 /mnt | |
670 | * mount /dev/sda3 /mnt | |
671 | * | |
672 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
673 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
674 | * /dev/sda2, then /dev/sda3, then NULL. | |
675 | * | |
676 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 677 | */ |
ca71cf71 | 678 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 679 | { |
c7105365 | 680 | struct mount *child_mnt; |
48a066e7 AV |
681 | struct vfsmount *m; |
682 | unsigned seq; | |
99b7db7b | 683 | |
48a066e7 AV |
684 | rcu_read_lock(); |
685 | do { | |
686 | seq = read_seqbegin(&mount_lock); | |
687 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
688 | m = child_mnt ? &child_mnt->mnt : NULL; | |
689 | } while (!legitimize_mnt(m, seq)); | |
690 | rcu_read_unlock(); | |
691 | return m; | |
a05964f3 RP |
692 | } |
693 | ||
7af1364f EB |
694 | /* |
695 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
696 | * current mount namespace. | |
697 | * | |
698 | * The common case is dentries are not mountpoints at all and that | |
699 | * test is handled inline. For the slow case when we are actually | |
700 | * dealing with a mountpoint of some kind, walk through all of the | |
701 | * mounts in the current mount namespace and test to see if the dentry | |
702 | * is a mountpoint. | |
703 | * | |
704 | * The mount_hashtable is not usable in the context because we | |
705 | * need to identify all mounts that may be in the current mount | |
706 | * namespace not just a mount that happens to have some specified | |
707 | * parent mount. | |
708 | */ | |
709 | bool __is_local_mountpoint(struct dentry *dentry) | |
710 | { | |
711 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
712 | struct mount *mnt; | |
713 | bool is_covered = false; | |
714 | ||
715 | if (!d_mountpoint(dentry)) | |
716 | goto out; | |
717 | ||
718 | down_read(&namespace_sem); | |
719 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
720 | is_covered = (mnt->mnt_mountpoint == dentry); | |
721 | if (is_covered) | |
722 | break; | |
723 | } | |
724 | up_read(&namespace_sem); | |
725 | out: | |
726 | return is_covered; | |
727 | } | |
728 | ||
e2dfa935 | 729 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 730 | { |
0818bf27 | 731 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
732 | struct mountpoint *mp; |
733 | ||
0818bf27 | 734 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
735 | if (mp->m_dentry == dentry) { |
736 | /* might be worth a WARN_ON() */ | |
737 | if (d_unlinked(dentry)) | |
738 | return ERR_PTR(-ENOENT); | |
739 | mp->m_count++; | |
740 | return mp; | |
741 | } | |
742 | } | |
e2dfa935 EB |
743 | return NULL; |
744 | } | |
745 | ||
3895dbf8 | 746 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 747 | { |
3895dbf8 | 748 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 749 | int ret; |
84d17192 | 750 | |
3895dbf8 EB |
751 | if (d_mountpoint(dentry)) { |
752 | mountpoint: | |
753 | read_seqlock_excl(&mount_lock); | |
754 | mp = lookup_mountpoint(dentry); | |
755 | read_sequnlock_excl(&mount_lock); | |
756 | if (mp) | |
757 | goto done; | |
758 | } | |
759 | ||
760 | if (!new) | |
761 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
762 | if (!new) | |
84d17192 AV |
763 | return ERR_PTR(-ENOMEM); |
764 | ||
3895dbf8 EB |
765 | |
766 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 767 | ret = d_set_mounted(dentry); |
eed81007 | 768 | |
3895dbf8 EB |
769 | /* Someone else set d_mounted? */ |
770 | if (ret == -EBUSY) | |
771 | goto mountpoint; | |
772 | ||
773 | /* The dentry is not available as a mountpoint? */ | |
774 | mp = ERR_PTR(ret); | |
775 | if (ret) | |
776 | goto done; | |
777 | ||
778 | /* Add the new mountpoint to the hash table */ | |
779 | read_seqlock_excl(&mount_lock); | |
780 | new->m_dentry = dentry; | |
781 | new->m_count = 1; | |
782 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
783 | INIT_HLIST_HEAD(&new->m_list); | |
784 | read_sequnlock_excl(&mount_lock); | |
785 | ||
786 | mp = new; | |
787 | new = NULL; | |
788 | done: | |
789 | kfree(new); | |
84d17192 AV |
790 | return mp; |
791 | } | |
792 | ||
793 | static void put_mountpoint(struct mountpoint *mp) | |
794 | { | |
795 | if (!--mp->m_count) { | |
796 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 797 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
798 | spin_lock(&dentry->d_lock); |
799 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
800 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 801 | hlist_del(&mp->m_hash); |
84d17192 AV |
802 | kfree(mp); |
803 | } | |
804 | } | |
805 | ||
143c8c91 | 806 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 807 | { |
6b3286ed | 808 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
809 | } |
810 | ||
99b7db7b NP |
811 | /* |
812 | * vfsmount lock must be held for write | |
813 | */ | |
6b3286ed | 814 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
815 | { |
816 | if (ns) { | |
817 | ns->event = ++event; | |
818 | wake_up_interruptible(&ns->poll); | |
819 | } | |
820 | } | |
821 | ||
99b7db7b NP |
822 | /* |
823 | * vfsmount lock must be held for write | |
824 | */ | |
6b3286ed | 825 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
826 | { |
827 | if (ns && ns->event != event) { | |
828 | ns->event = event; | |
829 | wake_up_interruptible(&ns->poll); | |
830 | } | |
831 | } | |
832 | ||
99b7db7b NP |
833 | /* |
834 | * vfsmount lock must be held for write | |
835 | */ | |
7bdb11de | 836 | static void unhash_mnt(struct mount *mnt) |
419148da | 837 | { |
0714a533 | 838 | mnt->mnt_parent = mnt; |
a73324da | 839 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 840 | list_del_init(&mnt->mnt_child); |
38129a13 | 841 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 842 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
843 | put_mountpoint(mnt->mnt_mp); |
844 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
845 | } |
846 | ||
7bdb11de EB |
847 | /* |
848 | * vfsmount lock must be held for write | |
849 | */ | |
850 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
851 | { | |
852 | old_path->dentry = mnt->mnt_mountpoint; | |
853 | old_path->mnt = &mnt->mnt_parent->mnt; | |
854 | unhash_mnt(mnt); | |
855 | } | |
856 | ||
6a46c573 EB |
857 | /* |
858 | * vfsmount lock must be held for write | |
859 | */ | |
860 | static void umount_mnt(struct mount *mnt) | |
861 | { | |
862 | /* old mountpoint will be dropped when we can do that */ | |
863 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
864 | unhash_mnt(mnt); | |
865 | } | |
866 | ||
99b7db7b NP |
867 | /* |
868 | * vfsmount lock must be held for write | |
869 | */ | |
84d17192 AV |
870 | void mnt_set_mountpoint(struct mount *mnt, |
871 | struct mountpoint *mp, | |
44d964d6 | 872 | struct mount *child_mnt) |
b90fa9ae | 873 | { |
84d17192 | 874 | mp->m_count++; |
3a2393d7 | 875 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 876 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 877 | child_mnt->mnt_parent = mnt; |
84d17192 | 878 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 879 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
880 | } |
881 | ||
99b7db7b NP |
882 | /* |
883 | * vfsmount lock must be held for write | |
884 | */ | |
84d17192 AV |
885 | static void attach_mnt(struct mount *mnt, |
886 | struct mount *parent, | |
887 | struct mountpoint *mp) | |
1da177e4 | 888 | { |
84d17192 | 889 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 890 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 891 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
892 | } |
893 | ||
12a5b529 AV |
894 | static void attach_shadowed(struct mount *mnt, |
895 | struct mount *parent, | |
896 | struct mount *shadows) | |
897 | { | |
898 | if (shadows) { | |
f6f99332 | 899 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
900 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
901 | } else { | |
902 | hlist_add_head_rcu(&mnt->mnt_hash, | |
903 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
904 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
905 | } | |
906 | } | |
907 | ||
b90fa9ae | 908 | /* |
99b7db7b | 909 | * vfsmount lock must be held for write |
b90fa9ae | 910 | */ |
1d6a32ac | 911 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 912 | { |
0714a533 | 913 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 914 | struct mount *m; |
b90fa9ae | 915 | LIST_HEAD(head); |
143c8c91 | 916 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 917 | |
0714a533 | 918 | BUG_ON(parent == mnt); |
b90fa9ae | 919 | |
1a4eeaf2 | 920 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 921 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 922 | m->mnt_ns = n; |
f03c6599 | 923 | |
b90fa9ae RP |
924 | list_splice(&head, n->list.prev); |
925 | ||
d2921684 EB |
926 | n->mounts += n->pending_mounts; |
927 | n->pending_mounts = 0; | |
928 | ||
12a5b529 | 929 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 930 | touch_mnt_namespace(n); |
1da177e4 LT |
931 | } |
932 | ||
909b0a88 | 933 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 934 | { |
6b41d536 AV |
935 | struct list_head *next = p->mnt_mounts.next; |
936 | if (next == &p->mnt_mounts) { | |
1da177e4 | 937 | while (1) { |
909b0a88 | 938 | if (p == root) |
1da177e4 | 939 | return NULL; |
6b41d536 AV |
940 | next = p->mnt_child.next; |
941 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 942 | break; |
0714a533 | 943 | p = p->mnt_parent; |
1da177e4 LT |
944 | } |
945 | } | |
6b41d536 | 946 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
947 | } |
948 | ||
315fc83e | 949 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 950 | { |
6b41d536 AV |
951 | struct list_head *prev = p->mnt_mounts.prev; |
952 | while (prev != &p->mnt_mounts) { | |
953 | p = list_entry(prev, struct mount, mnt_child); | |
954 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
955 | } |
956 | return p; | |
957 | } | |
958 | ||
9d412a43 AV |
959 | struct vfsmount * |
960 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
961 | { | |
b105e270 | 962 | struct mount *mnt; |
9d412a43 AV |
963 | struct dentry *root; |
964 | ||
965 | if (!type) | |
966 | return ERR_PTR(-ENODEV); | |
967 | ||
968 | mnt = alloc_vfsmnt(name); | |
969 | if (!mnt) | |
970 | return ERR_PTR(-ENOMEM); | |
971 | ||
972 | if (flags & MS_KERNMOUNT) | |
b105e270 | 973 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
974 | |
975 | root = mount_fs(type, flags, name, data); | |
976 | if (IS_ERR(root)) { | |
8ffcb32e | 977 | mnt_free_id(mnt); |
9d412a43 AV |
978 | free_vfsmnt(mnt); |
979 | return ERR_CAST(root); | |
980 | } | |
981 | ||
b105e270 AV |
982 | mnt->mnt.mnt_root = root; |
983 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 984 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 985 | mnt->mnt_parent = mnt; |
719ea2fb | 986 | lock_mount_hash(); |
39f7c4db | 987 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 988 | unlock_mount_hash(); |
b105e270 | 989 | return &mnt->mnt; |
9d412a43 AV |
990 | } |
991 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
992 | ||
c8d3e3d5 EB |
993 | struct vfsmount * |
994 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
995 | const char *name, void *data) | |
996 | { | |
997 | /* Until it is worked out how to pass the user namespace | |
998 | * through from the parent mount to the submount don't support | |
999 | * unprivileged mounts with submounts. | |
1000 | */ | |
1001 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1002 | return ERR_PTR(-EPERM); | |
1003 | ||
1004 | return vfs_kern_mount(type, MS_SUBMOUNT, name, data); | |
1005 | } | |
1006 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1007 | ||
87129cc0 | 1008 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1009 | int flag) |
1da177e4 | 1010 | { |
87129cc0 | 1011 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1012 | struct mount *mnt; |
1013 | int err; | |
1da177e4 | 1014 | |
be34d1a3 DH |
1015 | mnt = alloc_vfsmnt(old->mnt_devname); |
1016 | if (!mnt) | |
1017 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1018 | |
7a472ef4 | 1019 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1020 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1021 | else | |
1022 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1023 | |
be34d1a3 DH |
1024 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1025 | err = mnt_alloc_group_id(mnt); | |
1026 | if (err) | |
1027 | goto out_free; | |
1da177e4 | 1028 | } |
be34d1a3 | 1029 | |
f2ebb3a9 | 1030 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 1031 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1032 | if (flag & CL_UNPRIVILEGED) { |
1033 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1034 | ||
1035 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1036 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1037 | ||
1038 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1039 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1040 | ||
1041 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1042 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1043 | ||
1044 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1045 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1046 | } | |
132c94e3 | 1047 | |
5ff9d8a6 | 1048 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1049 | if ((flag & CL_UNPRIVILEGED) && |
1050 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1051 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1052 | ||
be34d1a3 DH |
1053 | atomic_inc(&sb->s_active); |
1054 | mnt->mnt.mnt_sb = sb; | |
1055 | mnt->mnt.mnt_root = dget(root); | |
1056 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1057 | mnt->mnt_parent = mnt; | |
719ea2fb | 1058 | lock_mount_hash(); |
be34d1a3 | 1059 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1060 | unlock_mount_hash(); |
be34d1a3 | 1061 | |
7a472ef4 EB |
1062 | if ((flag & CL_SLAVE) || |
1063 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1064 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1065 | mnt->mnt_master = old; | |
1066 | CLEAR_MNT_SHARED(mnt); | |
1067 | } else if (!(flag & CL_PRIVATE)) { | |
1068 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1069 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1070 | if (IS_MNT_SLAVE(old)) | |
1071 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1072 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1073 | } else { |
1074 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1075 | } |
1076 | if (flag & CL_MAKE_SHARED) | |
1077 | set_mnt_shared(mnt); | |
1078 | ||
1079 | /* stick the duplicate mount on the same expiry list | |
1080 | * as the original if that was on one */ | |
1081 | if (flag & CL_EXPIRE) { | |
1082 | if (!list_empty(&old->mnt_expire)) | |
1083 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1084 | } | |
1085 | ||
cb338d06 | 1086 | return mnt; |
719f5d7f MS |
1087 | |
1088 | out_free: | |
8ffcb32e | 1089 | mnt_free_id(mnt); |
719f5d7f | 1090 | free_vfsmnt(mnt); |
be34d1a3 | 1091 | return ERR_PTR(err); |
1da177e4 LT |
1092 | } |
1093 | ||
9ea459e1 AV |
1094 | static void cleanup_mnt(struct mount *mnt) |
1095 | { | |
1096 | /* | |
1097 | * This probably indicates that somebody messed | |
1098 | * up a mnt_want/drop_write() pair. If this | |
1099 | * happens, the filesystem was probably unable | |
1100 | * to make r/w->r/o transitions. | |
1101 | */ | |
1102 | /* | |
1103 | * The locking used to deal with mnt_count decrement provides barriers, | |
1104 | * so mnt_get_writers() below is safe. | |
1105 | */ | |
1106 | WARN_ON(mnt_get_writers(mnt)); | |
1107 | if (unlikely(mnt->mnt_pins.first)) | |
1108 | mnt_pin_kill(mnt); | |
1109 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1110 | dput(mnt->mnt.mnt_root); | |
1111 | deactivate_super(mnt->mnt.mnt_sb); | |
1112 | mnt_free_id(mnt); | |
1113 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1114 | } | |
1115 | ||
1116 | static void __cleanup_mnt(struct rcu_head *head) | |
1117 | { | |
1118 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1119 | } | |
1120 | ||
1121 | static LLIST_HEAD(delayed_mntput_list); | |
1122 | static void delayed_mntput(struct work_struct *unused) | |
1123 | { | |
1124 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1125 | struct llist_node *next; | |
1126 | ||
1127 | for (; node; node = next) { | |
1128 | next = llist_next(node); | |
1129 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1130 | } | |
1131 | } | |
1132 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1133 | ||
900148dc | 1134 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1135 | { |
48a066e7 AV |
1136 | rcu_read_lock(); |
1137 | mnt_add_count(mnt, -1); | |
1138 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1139 | rcu_read_unlock(); | |
f03c6599 | 1140 | return; |
b3e19d92 | 1141 | } |
719ea2fb | 1142 | lock_mount_hash(); |
b3e19d92 | 1143 | if (mnt_get_count(mnt)) { |
48a066e7 | 1144 | rcu_read_unlock(); |
719ea2fb | 1145 | unlock_mount_hash(); |
99b7db7b NP |
1146 | return; |
1147 | } | |
48a066e7 AV |
1148 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1149 | rcu_read_unlock(); | |
1150 | unlock_mount_hash(); | |
1151 | return; | |
1152 | } | |
1153 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1154 | rcu_read_unlock(); | |
962830df | 1155 | |
39f7c4db | 1156 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1157 | |
1158 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1159 | struct mount *p, *tmp; | |
1160 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1161 | umount_mnt(p); | |
1162 | } | |
1163 | } | |
719ea2fb | 1164 | unlock_mount_hash(); |
649a795a | 1165 | |
9ea459e1 AV |
1166 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1167 | struct task_struct *task = current; | |
1168 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1169 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1170 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1171 | return; | |
1172 | } | |
1173 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1174 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1175 | return; | |
1176 | } | |
1177 | cleanup_mnt(mnt); | |
b3e19d92 | 1178 | } |
b3e19d92 NP |
1179 | |
1180 | void mntput(struct vfsmount *mnt) | |
1181 | { | |
1182 | if (mnt) { | |
863d684f | 1183 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1184 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1185 | if (unlikely(m->mnt_expiry_mark)) |
1186 | m->mnt_expiry_mark = 0; | |
1187 | mntput_no_expire(m); | |
b3e19d92 NP |
1188 | } |
1189 | } | |
1190 | EXPORT_SYMBOL(mntput); | |
1191 | ||
1192 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1193 | { | |
1194 | if (mnt) | |
83adc753 | 1195 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1196 | return mnt; |
1197 | } | |
1198 | EXPORT_SYMBOL(mntget); | |
1199 | ||
c6609c0a IK |
1200 | /* path_is_mountpoint() - Check if path is a mount in the current |
1201 | * namespace. | |
1202 | * | |
1203 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1204 | * not mounted in any namespace and that common case is handled inline. | |
1205 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1206 | * mounts using a given dentry in a different namespace. This function | |
1207 | * checks if the passed in path is a mountpoint rather than the dentry | |
1208 | * alone. | |
1209 | */ | |
1210 | bool path_is_mountpoint(const struct path *path) | |
1211 | { | |
1212 | unsigned seq; | |
1213 | bool res; | |
1214 | ||
1215 | if (!d_mountpoint(path->dentry)) | |
1216 | return false; | |
1217 | ||
1218 | rcu_read_lock(); | |
1219 | do { | |
1220 | seq = read_seqbegin(&mount_lock); | |
1221 | res = __path_is_mountpoint(path); | |
1222 | } while (read_seqretry(&mount_lock, seq)); | |
1223 | rcu_read_unlock(); | |
1224 | ||
1225 | return res; | |
1226 | } | |
1227 | EXPORT_SYMBOL(path_is_mountpoint); | |
1228 | ||
ca71cf71 | 1229 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1230 | { |
3064c356 AV |
1231 | struct mount *p; |
1232 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1233 | if (IS_ERR(p)) | |
1234 | return ERR_CAST(p); | |
1235 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1236 | return &p->mnt; | |
7b7b1ace | 1237 | } |
1da177e4 | 1238 | |
b3b304a2 MS |
1239 | static inline void mangle(struct seq_file *m, const char *s) |
1240 | { | |
1241 | seq_escape(m, s, " \t\n\\"); | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * Simple .show_options callback for filesystems which don't want to | |
1246 | * implement more complex mount option showing. | |
1247 | * | |
1248 | * See also save_mount_options(). | |
1249 | */ | |
34c80b1d | 1250 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1251 | { |
2a32cebd AV |
1252 | const char *options; |
1253 | ||
1254 | rcu_read_lock(); | |
34c80b1d | 1255 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1256 | |
1257 | if (options != NULL && options[0]) { | |
1258 | seq_putc(m, ','); | |
1259 | mangle(m, options); | |
1260 | } | |
2a32cebd | 1261 | rcu_read_unlock(); |
b3b304a2 MS |
1262 | |
1263 | return 0; | |
1264 | } | |
1265 | EXPORT_SYMBOL(generic_show_options); | |
1266 | ||
1267 | /* | |
1268 | * If filesystem uses generic_show_options(), this function should be | |
1269 | * called from the fill_super() callback. | |
1270 | * | |
1271 | * The .remount_fs callback usually needs to be handled in a special | |
1272 | * way, to make sure, that previous options are not overwritten if the | |
1273 | * remount fails. | |
1274 | * | |
1275 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1276 | * reset all options to their default value, but changes only newly | |
1277 | * given options, then the displayed options will not reflect reality | |
1278 | * any more. | |
1279 | */ | |
1280 | void save_mount_options(struct super_block *sb, char *options) | |
1281 | { | |
2a32cebd AV |
1282 | BUG_ON(sb->s_options); |
1283 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1284 | } |
1285 | EXPORT_SYMBOL(save_mount_options); | |
1286 | ||
2a32cebd AV |
1287 | void replace_mount_options(struct super_block *sb, char *options) |
1288 | { | |
1289 | char *old = sb->s_options; | |
1290 | rcu_assign_pointer(sb->s_options, options); | |
1291 | if (old) { | |
1292 | synchronize_rcu(); | |
1293 | kfree(old); | |
1294 | } | |
1295 | } | |
1296 | EXPORT_SYMBOL(replace_mount_options); | |
1297 | ||
a1a2c409 | 1298 | #ifdef CONFIG_PROC_FS |
0226f492 | 1299 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1300 | static void *m_start(struct seq_file *m, loff_t *pos) |
1301 | { | |
ede1bf0d | 1302 | struct proc_mounts *p = m->private; |
1da177e4 | 1303 | |
390c6843 | 1304 | down_read(&namespace_sem); |
c7999c36 AV |
1305 | if (p->cached_event == p->ns->event) { |
1306 | void *v = p->cached_mount; | |
1307 | if (*pos == p->cached_index) | |
1308 | return v; | |
1309 | if (*pos == p->cached_index + 1) { | |
1310 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1311 | return p->cached_mount = v; | |
1312 | } | |
1313 | } | |
1314 | ||
1315 | p->cached_event = p->ns->event; | |
1316 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1317 | p->cached_index = *pos; | |
1318 | return p->cached_mount; | |
1da177e4 LT |
1319 | } |
1320 | ||
1321 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1322 | { | |
ede1bf0d | 1323 | struct proc_mounts *p = m->private; |
b0765fb8 | 1324 | |
c7999c36 AV |
1325 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1326 | p->cached_index = *pos; | |
1327 | return p->cached_mount; | |
1da177e4 LT |
1328 | } |
1329 | ||
1330 | static void m_stop(struct seq_file *m, void *v) | |
1331 | { | |
390c6843 | 1332 | up_read(&namespace_sem); |
1da177e4 LT |
1333 | } |
1334 | ||
0226f492 | 1335 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1336 | { |
ede1bf0d | 1337 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1338 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1339 | return p->show(m, &r->mnt); |
1da177e4 LT |
1340 | } |
1341 | ||
a1a2c409 | 1342 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1343 | .start = m_start, |
1344 | .next = m_next, | |
1345 | .stop = m_stop, | |
0226f492 | 1346 | .show = m_show, |
b4629fe2 | 1347 | }; |
a1a2c409 | 1348 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1349 | |
1da177e4 LT |
1350 | /** |
1351 | * may_umount_tree - check if a mount tree is busy | |
1352 | * @mnt: root of mount tree | |
1353 | * | |
1354 | * This is called to check if a tree of mounts has any | |
1355 | * open files, pwds, chroots or sub mounts that are | |
1356 | * busy. | |
1357 | */ | |
909b0a88 | 1358 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1359 | { |
909b0a88 | 1360 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1361 | int actual_refs = 0; |
1362 | int minimum_refs = 0; | |
315fc83e | 1363 | struct mount *p; |
909b0a88 | 1364 | BUG_ON(!m); |
1da177e4 | 1365 | |
b3e19d92 | 1366 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1367 | lock_mount_hash(); |
909b0a88 | 1368 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1369 | actual_refs += mnt_get_count(p); |
1da177e4 | 1370 | minimum_refs += 2; |
1da177e4 | 1371 | } |
719ea2fb | 1372 | unlock_mount_hash(); |
1da177e4 LT |
1373 | |
1374 | if (actual_refs > minimum_refs) | |
e3474a8e | 1375 | return 0; |
1da177e4 | 1376 | |
e3474a8e | 1377 | return 1; |
1da177e4 LT |
1378 | } |
1379 | ||
1380 | EXPORT_SYMBOL(may_umount_tree); | |
1381 | ||
1382 | /** | |
1383 | * may_umount - check if a mount point is busy | |
1384 | * @mnt: root of mount | |
1385 | * | |
1386 | * This is called to check if a mount point has any | |
1387 | * open files, pwds, chroots or sub mounts. If the | |
1388 | * mount has sub mounts this will return busy | |
1389 | * regardless of whether the sub mounts are busy. | |
1390 | * | |
1391 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1392 | * give false negatives. The main reason why it's here is that we need | |
1393 | * a non-destructive way to look for easily umountable filesystems. | |
1394 | */ | |
1395 | int may_umount(struct vfsmount *mnt) | |
1396 | { | |
e3474a8e | 1397 | int ret = 1; |
8ad08d8a | 1398 | down_read(&namespace_sem); |
719ea2fb | 1399 | lock_mount_hash(); |
1ab59738 | 1400 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1401 | ret = 0; |
719ea2fb | 1402 | unlock_mount_hash(); |
8ad08d8a | 1403 | up_read(&namespace_sem); |
a05964f3 | 1404 | return ret; |
1da177e4 LT |
1405 | } |
1406 | ||
1407 | EXPORT_SYMBOL(may_umount); | |
1408 | ||
38129a13 | 1409 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1410 | |
97216be0 | 1411 | static void namespace_unlock(void) |
70fbcdf4 | 1412 | { |
a3b3c562 | 1413 | struct hlist_head head; |
97216be0 | 1414 | |
a3b3c562 | 1415 | hlist_move_list(&unmounted, &head); |
97216be0 | 1416 | |
97216be0 AV |
1417 | up_write(&namespace_sem); |
1418 | ||
a3b3c562 EB |
1419 | if (likely(hlist_empty(&head))) |
1420 | return; | |
1421 | ||
48a066e7 AV |
1422 | synchronize_rcu(); |
1423 | ||
87b95ce0 | 1424 | group_pin_kill(&head); |
70fbcdf4 RP |
1425 | } |
1426 | ||
97216be0 | 1427 | static inline void namespace_lock(void) |
e3197d83 | 1428 | { |
97216be0 | 1429 | down_write(&namespace_sem); |
e3197d83 AV |
1430 | } |
1431 | ||
e819f152 EB |
1432 | enum umount_tree_flags { |
1433 | UMOUNT_SYNC = 1, | |
1434 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1435 | UMOUNT_CONNECTED = 4, |
e819f152 | 1436 | }; |
f2d0a123 EB |
1437 | |
1438 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1439 | { | |
1440 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1441 | if (how & UMOUNT_SYNC) | |
1442 | return true; | |
1443 | ||
1444 | /* A mount without a parent has nothing to be connected to */ | |
1445 | if (!mnt_has_parent(mnt)) | |
1446 | return true; | |
1447 | ||
1448 | /* Because the reference counting rules change when mounts are | |
1449 | * unmounted and connected, umounted mounts may not be | |
1450 | * connected to mounted mounts. | |
1451 | */ | |
1452 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1453 | return true; | |
1454 | ||
1455 | /* Has it been requested that the mount remain connected? */ | |
1456 | if (how & UMOUNT_CONNECTED) | |
1457 | return false; | |
1458 | ||
1459 | /* Is the mount locked such that it needs to remain connected? */ | |
1460 | if (IS_MNT_LOCKED(mnt)) | |
1461 | return false; | |
1462 | ||
1463 | /* By default disconnect the mount */ | |
1464 | return true; | |
1465 | } | |
1466 | ||
99b7db7b | 1467 | /* |
48a066e7 | 1468 | * mount_lock must be held |
99b7db7b NP |
1469 | * namespace_sem must be held for write |
1470 | */ | |
e819f152 | 1471 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1472 | { |
c003b26f | 1473 | LIST_HEAD(tmp_list); |
315fc83e | 1474 | struct mount *p; |
1da177e4 | 1475 | |
5d88457e EB |
1476 | if (how & UMOUNT_PROPAGATE) |
1477 | propagate_mount_unlock(mnt); | |
1478 | ||
c003b26f | 1479 | /* Gather the mounts to umount */ |
590ce4bc EB |
1480 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1481 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1482 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1483 | } |
1da177e4 | 1484 | |
411a938b | 1485 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1486 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1487 | list_del_init(&p->mnt_child); |
c003b26f | 1488 | } |
88b368f2 | 1489 | |
c003b26f | 1490 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1491 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1492 | propagate_umount(&tmp_list); |
a05964f3 | 1493 | |
c003b26f | 1494 | while (!list_empty(&tmp_list)) { |
d2921684 | 1495 | struct mnt_namespace *ns; |
ce07d891 | 1496 | bool disconnect; |
c003b26f | 1497 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1498 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1499 | list_del_init(&p->mnt_list); |
d2921684 EB |
1500 | ns = p->mnt_ns; |
1501 | if (ns) { | |
1502 | ns->mounts--; | |
1503 | __touch_mnt_namespace(ns); | |
1504 | } | |
143c8c91 | 1505 | p->mnt_ns = NULL; |
e819f152 | 1506 | if (how & UMOUNT_SYNC) |
48a066e7 | 1507 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1508 | |
f2d0a123 | 1509 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1510 | |
1511 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1512 | disconnect ? &unmounted : NULL); | |
676da58d | 1513 | if (mnt_has_parent(p)) { |
81b6b061 | 1514 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1515 | if (!disconnect) { |
1516 | /* Don't forget about p */ | |
1517 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1518 | } else { | |
1519 | umount_mnt(p); | |
1520 | } | |
7c4b93d8 | 1521 | } |
0f0afb1d | 1522 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1523 | } |
1524 | } | |
1525 | ||
b54b9be7 | 1526 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1527 | |
1ab59738 | 1528 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1529 | { |
1ab59738 | 1530 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1531 | int retval; |
1532 | ||
1ab59738 | 1533 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1534 | if (retval) |
1535 | return retval; | |
1536 | ||
1537 | /* | |
1538 | * Allow userspace to request a mountpoint be expired rather than | |
1539 | * unmounting unconditionally. Unmount only happens if: | |
1540 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1541 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1542 | */ | |
1543 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1544 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1545 | flags & (MNT_FORCE | MNT_DETACH)) |
1546 | return -EINVAL; | |
1547 | ||
b3e19d92 NP |
1548 | /* |
1549 | * probably don't strictly need the lock here if we examined | |
1550 | * all race cases, but it's a slowpath. | |
1551 | */ | |
719ea2fb | 1552 | lock_mount_hash(); |
83adc753 | 1553 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1554 | unlock_mount_hash(); |
1da177e4 | 1555 | return -EBUSY; |
b3e19d92 | 1556 | } |
719ea2fb | 1557 | unlock_mount_hash(); |
1da177e4 | 1558 | |
863d684f | 1559 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1560 | return -EAGAIN; |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * If we may have to abort operations to get out of this | |
1565 | * mount, and they will themselves hold resources we must | |
1566 | * allow the fs to do things. In the Unix tradition of | |
1567 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1568 | * might fail to complete on the first run through as other tasks | |
1569 | * must return, and the like. Thats for the mount program to worry | |
1570 | * about for the moment. | |
1571 | */ | |
1572 | ||
42faad99 | 1573 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1574 | sb->s_op->umount_begin(sb); |
42faad99 | 1575 | } |
1da177e4 LT |
1576 | |
1577 | /* | |
1578 | * No sense to grab the lock for this test, but test itself looks | |
1579 | * somewhat bogus. Suggestions for better replacement? | |
1580 | * Ho-hum... In principle, we might treat that as umount + switch | |
1581 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1582 | * Actually it makes sense, especially if rootfs would contain a | |
1583 | * /reboot - static binary that would close all descriptors and | |
1584 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1585 | */ | |
1ab59738 | 1586 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1587 | /* |
1588 | * Special case for "unmounting" root ... | |
1589 | * we just try to remount it readonly. | |
1590 | */ | |
f17a3091 | 1591 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1592 | return -EPERM; |
1da177e4 | 1593 | down_write(&sb->s_umount); |
4aa98cf7 | 1594 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1595 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1596 | up_write(&sb->s_umount); |
1597 | return retval; | |
1598 | } | |
1599 | ||
97216be0 | 1600 | namespace_lock(); |
719ea2fb | 1601 | lock_mount_hash(); |
5addc5dd | 1602 | event++; |
1da177e4 | 1603 | |
48a066e7 | 1604 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1605 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1606 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1607 | retval = 0; |
48a066e7 AV |
1608 | } else { |
1609 | shrink_submounts(mnt); | |
1610 | retval = -EBUSY; | |
1611 | if (!propagate_mount_busy(mnt, 2)) { | |
1612 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1613 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1614 | retval = 0; |
1615 | } | |
1da177e4 | 1616 | } |
719ea2fb | 1617 | unlock_mount_hash(); |
e3197d83 | 1618 | namespace_unlock(); |
1da177e4 LT |
1619 | return retval; |
1620 | } | |
1621 | ||
80b5dce8 EB |
1622 | /* |
1623 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1624 | * | |
1625 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1626 | * to an existing mountpoint, and wind up leaking the mount. | |
1627 | * detach_mounts allows lazily unmounting those mounts instead of | |
1628 | * leaking them. | |
1629 | * | |
1630 | * The caller may hold dentry->d_inode->i_mutex. | |
1631 | */ | |
1632 | void __detach_mounts(struct dentry *dentry) | |
1633 | { | |
1634 | struct mountpoint *mp; | |
1635 | struct mount *mnt; | |
1636 | ||
1637 | namespace_lock(); | |
3895dbf8 | 1638 | lock_mount_hash(); |
80b5dce8 | 1639 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1640 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1641 | goto out_unlock; |
1642 | ||
e06b933e | 1643 | event++; |
80b5dce8 EB |
1644 | while (!hlist_empty(&mp->m_list)) { |
1645 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1646 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1647 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1648 | umount_mnt(mnt); | |
ce07d891 | 1649 | } |
e0c9c0af | 1650 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1651 | } |
80b5dce8 EB |
1652 | put_mountpoint(mp); |
1653 | out_unlock: | |
3895dbf8 | 1654 | unlock_mount_hash(); |
80b5dce8 EB |
1655 | namespace_unlock(); |
1656 | } | |
1657 | ||
9b40bc90 AV |
1658 | /* |
1659 | * Is the caller allowed to modify his namespace? | |
1660 | */ | |
1661 | static inline bool may_mount(void) | |
1662 | { | |
1663 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1664 | } | |
1665 | ||
9e8925b6 JL |
1666 | static inline bool may_mandlock(void) |
1667 | { | |
1668 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1669 | return false; | |
1670 | #endif | |
95ace754 | 1671 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1672 | } |
1673 | ||
1da177e4 LT |
1674 | /* |
1675 | * Now umount can handle mount points as well as block devices. | |
1676 | * This is important for filesystems which use unnamed block devices. | |
1677 | * | |
1678 | * We now support a flag for forced unmount like the other 'big iron' | |
1679 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1680 | */ | |
1681 | ||
bdc480e3 | 1682 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1683 | { |
2d8f3038 | 1684 | struct path path; |
900148dc | 1685 | struct mount *mnt; |
1da177e4 | 1686 | int retval; |
db1f05bb | 1687 | int lookup_flags = 0; |
1da177e4 | 1688 | |
db1f05bb MS |
1689 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1690 | return -EINVAL; | |
1691 | ||
9b40bc90 AV |
1692 | if (!may_mount()) |
1693 | return -EPERM; | |
1694 | ||
db1f05bb MS |
1695 | if (!(flags & UMOUNT_NOFOLLOW)) |
1696 | lookup_flags |= LOOKUP_FOLLOW; | |
1697 | ||
197df04c | 1698 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1699 | if (retval) |
1700 | goto out; | |
900148dc | 1701 | mnt = real_mount(path.mnt); |
1da177e4 | 1702 | retval = -EINVAL; |
2d8f3038 | 1703 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1704 | goto dput_and_out; |
143c8c91 | 1705 | if (!check_mnt(mnt)) |
1da177e4 | 1706 | goto dput_and_out; |
5ff9d8a6 EB |
1707 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1708 | goto dput_and_out; | |
b2f5d4dc EB |
1709 | retval = -EPERM; |
1710 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1711 | goto dput_and_out; | |
1da177e4 | 1712 | |
900148dc | 1713 | retval = do_umount(mnt, flags); |
1da177e4 | 1714 | dput_and_out: |
429731b1 | 1715 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1716 | dput(path.dentry); |
900148dc | 1717 | mntput_no_expire(mnt); |
1da177e4 LT |
1718 | out: |
1719 | return retval; | |
1720 | } | |
1721 | ||
1722 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1723 | ||
1724 | /* | |
b58fed8b | 1725 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1726 | */ |
bdc480e3 | 1727 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1728 | { |
b58fed8b | 1729 | return sys_umount(name, 0); |
1da177e4 LT |
1730 | } |
1731 | ||
1732 | #endif | |
1733 | ||
4ce5d2b1 | 1734 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1735 | { |
4ce5d2b1 | 1736 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1737 | return dentry->d_op == &ns_dentry_operations && |
1738 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1739 | } |
1740 | ||
58be2825 AV |
1741 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1742 | { | |
1743 | return container_of(ns, struct mnt_namespace, ns); | |
1744 | } | |
1745 | ||
4ce5d2b1 EB |
1746 | static bool mnt_ns_loop(struct dentry *dentry) |
1747 | { | |
1748 | /* Could bind mounting the mount namespace inode cause a | |
1749 | * mount namespace loop? | |
1750 | */ | |
1751 | struct mnt_namespace *mnt_ns; | |
1752 | if (!is_mnt_ns_file(dentry)) | |
1753 | return false; | |
1754 | ||
f77c8014 | 1755 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1756 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1757 | } | |
1758 | ||
87129cc0 | 1759 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1760 | int flag) |
1da177e4 | 1761 | { |
84d17192 | 1762 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1763 | |
4ce5d2b1 EB |
1764 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1765 | return ERR_PTR(-EINVAL); | |
1766 | ||
1767 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1768 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1769 | |
36341f64 | 1770 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1771 | if (IS_ERR(q)) |
1772 | return q; | |
1773 | ||
a73324da | 1774 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1775 | |
1776 | p = mnt; | |
6b41d536 | 1777 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1778 | struct mount *s; |
7ec02ef1 | 1779 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1780 | continue; |
1781 | ||
909b0a88 | 1782 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1783 | struct mount *t = NULL; |
4ce5d2b1 EB |
1784 | if (!(flag & CL_COPY_UNBINDABLE) && |
1785 | IS_MNT_UNBINDABLE(s)) { | |
1786 | s = skip_mnt_tree(s); | |
1787 | continue; | |
1788 | } | |
1789 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1790 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1791 | s = skip_mnt_tree(s); |
1792 | continue; | |
1793 | } | |
0714a533 AV |
1794 | while (p != s->mnt_parent) { |
1795 | p = p->mnt_parent; | |
1796 | q = q->mnt_parent; | |
1da177e4 | 1797 | } |
87129cc0 | 1798 | p = s; |
84d17192 | 1799 | parent = q; |
87129cc0 | 1800 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1801 | if (IS_ERR(q)) |
1802 | goto out; | |
719ea2fb | 1803 | lock_mount_hash(); |
1a4eeaf2 | 1804 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1805 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1806 | if (!list_empty(&parent->mnt_mounts)) { | |
1807 | t = list_last_entry(&parent->mnt_mounts, | |
1808 | struct mount, mnt_child); | |
1809 | if (t->mnt_mp != p->mnt_mp) | |
1810 | t = NULL; | |
1811 | } | |
1812 | attach_shadowed(q, parent, t); | |
719ea2fb | 1813 | unlock_mount_hash(); |
1da177e4 LT |
1814 | } |
1815 | } | |
1816 | return res; | |
be34d1a3 | 1817 | out: |
1da177e4 | 1818 | if (res) { |
719ea2fb | 1819 | lock_mount_hash(); |
e819f152 | 1820 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1821 | unlock_mount_hash(); |
1da177e4 | 1822 | } |
be34d1a3 | 1823 | return q; |
1da177e4 LT |
1824 | } |
1825 | ||
be34d1a3 DH |
1826 | /* Caller should check returned pointer for errors */ |
1827 | ||
ca71cf71 | 1828 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1829 | { |
cb338d06 | 1830 | struct mount *tree; |
97216be0 | 1831 | namespace_lock(); |
cd4a4017 EB |
1832 | if (!check_mnt(real_mount(path->mnt))) |
1833 | tree = ERR_PTR(-EINVAL); | |
1834 | else | |
1835 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1836 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1837 | namespace_unlock(); |
be34d1a3 | 1838 | if (IS_ERR(tree)) |
52e220d3 | 1839 | return ERR_CAST(tree); |
be34d1a3 | 1840 | return &tree->mnt; |
8aec0809 AV |
1841 | } |
1842 | ||
1843 | void drop_collected_mounts(struct vfsmount *mnt) | |
1844 | { | |
97216be0 | 1845 | namespace_lock(); |
719ea2fb | 1846 | lock_mount_hash(); |
e819f152 | 1847 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1848 | unlock_mount_hash(); |
3ab6abee | 1849 | namespace_unlock(); |
8aec0809 AV |
1850 | } |
1851 | ||
c771d683 MS |
1852 | /** |
1853 | * clone_private_mount - create a private clone of a path | |
1854 | * | |
1855 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1856 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1857 | * to the originating mount won't be propagated into this). | |
1858 | * | |
1859 | * Release with mntput(). | |
1860 | */ | |
ca71cf71 | 1861 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1862 | { |
1863 | struct mount *old_mnt = real_mount(path->mnt); | |
1864 | struct mount *new_mnt; | |
1865 | ||
1866 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1867 | return ERR_PTR(-EINVAL); | |
1868 | ||
c771d683 | 1869 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1870 | if (IS_ERR(new_mnt)) |
1871 | return ERR_CAST(new_mnt); | |
1872 | ||
1873 | return &new_mnt->mnt; | |
1874 | } | |
1875 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1876 | ||
1f707137 AV |
1877 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1878 | struct vfsmount *root) | |
1879 | { | |
1a4eeaf2 | 1880 | struct mount *mnt; |
1f707137 AV |
1881 | int res = f(root, arg); |
1882 | if (res) | |
1883 | return res; | |
1a4eeaf2 AV |
1884 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1885 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1886 | if (res) |
1887 | return res; | |
1888 | } | |
1889 | return 0; | |
1890 | } | |
e14748e8 | 1891 | EXPORT_SYMBOL_GPL(iterate_mounts); |
1f707137 | 1892 | |
4b8b21f4 | 1893 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1894 | { |
315fc83e | 1895 | struct mount *p; |
719f5d7f | 1896 | |
909b0a88 | 1897 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1898 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1899 | mnt_release_group_id(p); |
719f5d7f MS |
1900 | } |
1901 | } | |
1902 | ||
4b8b21f4 | 1903 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1904 | { |
315fc83e | 1905 | struct mount *p; |
719f5d7f | 1906 | |
909b0a88 | 1907 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1908 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1909 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1910 | if (err) { |
4b8b21f4 | 1911 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1912 | return err; |
1913 | } | |
1914 | } | |
1915 | } | |
1916 | ||
1917 | return 0; | |
1918 | } | |
1919 | ||
d2921684 EB |
1920 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1921 | { | |
1922 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1923 | unsigned int mounts = 0, old, pending, sum; | |
1924 | struct mount *p; | |
1925 | ||
1926 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1927 | mounts++; | |
1928 | ||
1929 | old = ns->mounts; | |
1930 | pending = ns->pending_mounts; | |
1931 | sum = old + pending; | |
1932 | if ((old > sum) || | |
1933 | (pending > sum) || | |
1934 | (max < sum) || | |
1935 | (mounts > (max - sum))) | |
1936 | return -ENOSPC; | |
1937 | ||
1938 | ns->pending_mounts = pending + mounts; | |
1939 | return 0; | |
1940 | } | |
1941 | ||
b90fa9ae RP |
1942 | /* |
1943 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1944 | * @nd : place the mount tree @source_mnt is attached |
1945 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1946 | * store the parent mount and mountpoint dentry. | |
1947 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1948 | * |
1949 | * NOTE: in the table below explains the semantics when a source mount | |
1950 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1951 | * --------------------------------------------------------------------------- |
1952 | * | BIND MOUNT OPERATION | | |
1953 | * |************************************************************************** | |
1954 | * | source-->| shared | private | slave | unbindable | | |
1955 | * | dest | | | | | | |
1956 | * | | | | | | | | |
1957 | * | v | | | | | | |
1958 | * |************************************************************************** | |
1959 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1960 | * | | | | | | | |
1961 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1962 | * *************************************************************************** | |
b90fa9ae RP |
1963 | * A bind operation clones the source mount and mounts the clone on the |
1964 | * destination mount. | |
1965 | * | |
1966 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1967 | * tree of the destination mount and the cloned mount is added to | |
1968 | * the peer group of the source mount. | |
1969 | * (+) the cloned mount is created under the destination mount and is marked | |
1970 | * as shared. The cloned mount is added to the peer group of the source | |
1971 | * mount. | |
5afe0022 RP |
1972 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1973 | * of the destination mount and the cloned mount is made slave | |
1974 | * of the same master as that of the source mount. The cloned mount | |
1975 | * is marked as 'shared and slave'. | |
1976 | * (*) the cloned mount is made a slave of the same master as that of the | |
1977 | * source mount. | |
1978 | * | |
9676f0c6 RP |
1979 | * --------------------------------------------------------------------------- |
1980 | * | MOVE MOUNT OPERATION | | |
1981 | * |************************************************************************** | |
1982 | * | source-->| shared | private | slave | unbindable | | |
1983 | * | dest | | | | | | |
1984 | * | | | | | | | | |
1985 | * | v | | | | | | |
1986 | * |************************************************************************** | |
1987 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1988 | * | | | | | | | |
1989 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1990 | * *************************************************************************** | |
5afe0022 RP |
1991 | * |
1992 | * (+) the mount is moved to the destination. And is then propagated to | |
1993 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1994 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1995 | * (+++) the mount is moved to the destination and is then propagated to |
1996 | * all the mounts belonging to the destination mount's propagation tree. | |
1997 | * the mount is marked as 'shared and slave'. | |
1998 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1999 | * |
2000 | * if the source mount is a tree, the operations explained above is | |
2001 | * applied to each mount in the tree. | |
2002 | * Must be called without spinlocks held, since this function can sleep | |
2003 | * in allocations. | |
2004 | */ | |
0fb54e50 | 2005 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2006 | struct mount *dest_mnt, |
2007 | struct mountpoint *dest_mp, | |
2008 | struct path *parent_path) | |
b90fa9ae | 2009 | { |
38129a13 | 2010 | HLIST_HEAD(tree_list); |
d2921684 | 2011 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
315fc83e | 2012 | struct mount *child, *p; |
38129a13 | 2013 | struct hlist_node *n; |
719f5d7f | 2014 | int err; |
b90fa9ae | 2015 | |
d2921684 EB |
2016 | /* Is there space to add these mounts to the mount namespace? */ |
2017 | if (!parent_path) { | |
2018 | err = count_mounts(ns, source_mnt); | |
2019 | if (err) | |
2020 | goto out; | |
2021 | } | |
2022 | ||
fc7be130 | 2023 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2024 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2025 | if (err) |
2026 | goto out; | |
0b1b901b | 2027 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2028 | lock_mount_hash(); |
0b1b901b AV |
2029 | if (err) |
2030 | goto out_cleanup_ids; | |
909b0a88 | 2031 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2032 | set_mnt_shared(p); |
0b1b901b AV |
2033 | } else { |
2034 | lock_mount_hash(); | |
b90fa9ae | 2035 | } |
1a390689 | 2036 | if (parent_path) { |
0fb54e50 | 2037 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2038 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2039 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2040 | } else { |
84d17192 | 2041 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 2042 | commit_tree(source_mnt, NULL); |
21444403 | 2043 | } |
b90fa9ae | 2044 | |
38129a13 | 2045 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2046 | struct mount *q; |
38129a13 | 2047 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
2048 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
2049 | child->mnt_mountpoint); | |
2050 | commit_tree(child, q); | |
b90fa9ae | 2051 | } |
719ea2fb | 2052 | unlock_mount_hash(); |
99b7db7b | 2053 | |
b90fa9ae | 2054 | return 0; |
719f5d7f MS |
2055 | |
2056 | out_cleanup_ids: | |
f2ebb3a9 AV |
2057 | while (!hlist_empty(&tree_list)) { |
2058 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2059 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2060 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2061 | } |
2062 | unlock_mount_hash(); | |
0b1b901b | 2063 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2064 | out: |
d2921684 | 2065 | ns->pending_mounts = 0; |
719f5d7f | 2066 | return err; |
b90fa9ae RP |
2067 | } |
2068 | ||
84d17192 | 2069 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2070 | { |
2071 | struct vfsmount *mnt; | |
84d17192 | 2072 | struct dentry *dentry = path->dentry; |
b12cea91 | 2073 | retry: |
5955102c | 2074 | inode_lock(dentry->d_inode); |
84d17192 | 2075 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2076 | inode_unlock(dentry->d_inode); |
84d17192 | 2077 | return ERR_PTR(-ENOENT); |
b12cea91 | 2078 | } |
97216be0 | 2079 | namespace_lock(); |
b12cea91 | 2080 | mnt = lookup_mnt(path); |
84d17192 | 2081 | if (likely(!mnt)) { |
3895dbf8 | 2082 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2083 | if (IS_ERR(mp)) { |
97216be0 | 2084 | namespace_unlock(); |
5955102c | 2085 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2086 | return mp; |
2087 | } | |
2088 | return mp; | |
2089 | } | |
97216be0 | 2090 | namespace_unlock(); |
5955102c | 2091 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2092 | path_put(path); |
2093 | path->mnt = mnt; | |
84d17192 | 2094 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2095 | goto retry; |
2096 | } | |
2097 | ||
84d17192 | 2098 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2099 | { |
84d17192 | 2100 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2101 | |
2102 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2103 | put_mountpoint(where); |
3895dbf8 EB |
2104 | read_sequnlock_excl(&mount_lock); |
2105 | ||
328e6d90 | 2106 | namespace_unlock(); |
5955102c | 2107 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2108 | } |
2109 | ||
84d17192 | 2110 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2111 | { |
95bc5f25 | 2112 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
2113 | return -EINVAL; |
2114 | ||
e36cb0b8 DH |
2115 | if (d_is_dir(mp->m_dentry) != |
2116 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2117 | return -ENOTDIR; |
2118 | ||
84d17192 | 2119 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2120 | } |
2121 | ||
7a2e8a8f VA |
2122 | /* |
2123 | * Sanity check the flags to change_mnt_propagation. | |
2124 | */ | |
2125 | ||
2126 | static int flags_to_propagation_type(int flags) | |
2127 | { | |
7c6e984d | 2128 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2129 | |
2130 | /* Fail if any non-propagation flags are set */ | |
2131 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2132 | return 0; | |
2133 | /* Only one propagation flag should be set */ | |
2134 | if (!is_power_of_2(type)) | |
2135 | return 0; | |
2136 | return type; | |
2137 | } | |
2138 | ||
07b20889 RP |
2139 | /* |
2140 | * recursively change the type of the mountpoint. | |
2141 | */ | |
0a0d8a46 | 2142 | static int do_change_type(struct path *path, int flag) |
07b20889 | 2143 | { |
315fc83e | 2144 | struct mount *m; |
4b8b21f4 | 2145 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 2146 | int recurse = flag & MS_REC; |
7a2e8a8f | 2147 | int type; |
719f5d7f | 2148 | int err = 0; |
07b20889 | 2149 | |
2d92ab3c | 2150 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2151 | return -EINVAL; |
2152 | ||
7a2e8a8f VA |
2153 | type = flags_to_propagation_type(flag); |
2154 | if (!type) | |
2155 | return -EINVAL; | |
2156 | ||
97216be0 | 2157 | namespace_lock(); |
719f5d7f MS |
2158 | if (type == MS_SHARED) { |
2159 | err = invent_group_ids(mnt, recurse); | |
2160 | if (err) | |
2161 | goto out_unlock; | |
2162 | } | |
2163 | ||
719ea2fb | 2164 | lock_mount_hash(); |
909b0a88 | 2165 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2166 | change_mnt_propagation(m, type); |
719ea2fb | 2167 | unlock_mount_hash(); |
719f5d7f MS |
2168 | |
2169 | out_unlock: | |
97216be0 | 2170 | namespace_unlock(); |
719f5d7f | 2171 | return err; |
07b20889 RP |
2172 | } |
2173 | ||
5ff9d8a6 EB |
2174 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2175 | { | |
2176 | struct mount *child; | |
2177 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2178 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2179 | continue; | |
2180 | ||
2181 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2182 | return true; | |
2183 | } | |
2184 | return false; | |
2185 | } | |
2186 | ||
1da177e4 LT |
2187 | /* |
2188 | * do loopback mount. | |
2189 | */ | |
808d4e3c | 2190 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2191 | int recurse) |
1da177e4 | 2192 | { |
2d92ab3c | 2193 | struct path old_path; |
84d17192 AV |
2194 | struct mount *mnt = NULL, *old, *parent; |
2195 | struct mountpoint *mp; | |
57eccb83 | 2196 | int err; |
1da177e4 LT |
2197 | if (!old_name || !*old_name) |
2198 | return -EINVAL; | |
815d405c | 2199 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2200 | if (err) |
2201 | return err; | |
2202 | ||
8823c079 | 2203 | err = -EINVAL; |
4ce5d2b1 | 2204 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2205 | goto out; |
2206 | ||
84d17192 AV |
2207 | mp = lock_mount(path); |
2208 | err = PTR_ERR(mp); | |
2209 | if (IS_ERR(mp)) | |
b12cea91 AV |
2210 | goto out; |
2211 | ||
87129cc0 | 2212 | old = real_mount(old_path.mnt); |
84d17192 | 2213 | parent = real_mount(path->mnt); |
87129cc0 | 2214 | |
1da177e4 | 2215 | err = -EINVAL; |
fc7be130 | 2216 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2217 | goto out2; |
9676f0c6 | 2218 | |
e149ed2b AV |
2219 | if (!check_mnt(parent)) |
2220 | goto out2; | |
2221 | ||
2222 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2223 | goto out2; |
1da177e4 | 2224 | |
5ff9d8a6 EB |
2225 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2226 | goto out2; | |
2227 | ||
ccd48bc7 | 2228 | if (recurse) |
4ce5d2b1 | 2229 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2230 | else |
87129cc0 | 2231 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2232 | |
be34d1a3 DH |
2233 | if (IS_ERR(mnt)) { |
2234 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2235 | goto out2; |
be34d1a3 | 2236 | } |
ccd48bc7 | 2237 | |
5ff9d8a6 EB |
2238 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2239 | ||
84d17192 | 2240 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2241 | if (err) { |
719ea2fb | 2242 | lock_mount_hash(); |
e819f152 | 2243 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2244 | unlock_mount_hash(); |
5b83d2c5 | 2245 | } |
b12cea91 | 2246 | out2: |
84d17192 | 2247 | unlock_mount(mp); |
ccd48bc7 | 2248 | out: |
2d92ab3c | 2249 | path_put(&old_path); |
1da177e4 LT |
2250 | return err; |
2251 | } | |
2252 | ||
2e4b7fcd DH |
2253 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2254 | { | |
2255 | int error = 0; | |
2256 | int readonly_request = 0; | |
2257 | ||
2258 | if (ms_flags & MS_RDONLY) | |
2259 | readonly_request = 1; | |
2260 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2261 | return 0; | |
2262 | ||
2263 | if (readonly_request) | |
83adc753 | 2264 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2265 | else |
83adc753 | 2266 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2267 | return error; |
2268 | } | |
2269 | ||
1da177e4 LT |
2270 | /* |
2271 | * change filesystem flags. dir should be a physical root of filesystem. | |
2272 | * If you've mounted a non-root directory somewhere and want to do remount | |
2273 | * on it - tough luck. | |
2274 | */ | |
0a0d8a46 | 2275 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2276 | void *data) |
2277 | { | |
2278 | int err; | |
2d92ab3c | 2279 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2280 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2281 | |
143c8c91 | 2282 | if (!check_mnt(mnt)) |
1da177e4 LT |
2283 | return -EINVAL; |
2284 | ||
2d92ab3c | 2285 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2286 | return -EINVAL; |
2287 | ||
07b64558 EB |
2288 | /* Don't allow changing of locked mnt flags. |
2289 | * | |
2290 | * No locks need to be held here while testing the various | |
2291 | * MNT_LOCK flags because those flags can never be cleared | |
2292 | * once they are set. | |
2293 | */ | |
2294 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2295 | !(mnt_flags & MNT_READONLY)) { | |
2296 | return -EPERM; | |
2297 | } | |
9566d674 EB |
2298 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2299 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2300 | return -EPERM; |
9566d674 EB |
2301 | } |
2302 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2303 | !(mnt_flags & MNT_NOSUID)) { | |
2304 | return -EPERM; | |
2305 | } | |
2306 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2307 | !(mnt_flags & MNT_NOEXEC)) { | |
2308 | return -EPERM; | |
2309 | } | |
2310 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2311 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2312 | return -EPERM; | |
2313 | } | |
2314 | ||
ff36fe2c EP |
2315 | err = security_sb_remount(sb, data); |
2316 | if (err) | |
2317 | return err; | |
2318 | ||
1da177e4 | 2319 | down_write(&sb->s_umount); |
2e4b7fcd | 2320 | if (flags & MS_BIND) |
2d92ab3c | 2321 | err = change_mount_flags(path->mnt, flags); |
f17a3091 | 2322 | else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
57eccb83 | 2323 | err = -EPERM; |
4aa98cf7 | 2324 | else |
2e4b7fcd | 2325 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2326 | if (!err) { |
719ea2fb | 2327 | lock_mount_hash(); |
a6138db8 | 2328 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2329 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2330 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2331 | unlock_mount_hash(); |
0e55a7cc | 2332 | } |
6339dab8 | 2333 | up_write(&sb->s_umount); |
1da177e4 LT |
2334 | return err; |
2335 | } | |
2336 | ||
cbbe362c | 2337 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2338 | { |
315fc83e | 2339 | struct mount *p; |
909b0a88 | 2340 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2341 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2342 | return 1; |
2343 | } | |
2344 | return 0; | |
2345 | } | |
2346 | ||
808d4e3c | 2347 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2348 | { |
2d92ab3c | 2349 | struct path old_path, parent_path; |
676da58d | 2350 | struct mount *p; |
0fb54e50 | 2351 | struct mount *old; |
84d17192 | 2352 | struct mountpoint *mp; |
57eccb83 | 2353 | int err; |
1da177e4 LT |
2354 | if (!old_name || !*old_name) |
2355 | return -EINVAL; | |
2d92ab3c | 2356 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2357 | if (err) |
2358 | return err; | |
2359 | ||
84d17192 AV |
2360 | mp = lock_mount(path); |
2361 | err = PTR_ERR(mp); | |
2362 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2363 | goto out; |
2364 | ||
143c8c91 | 2365 | old = real_mount(old_path.mnt); |
fc7be130 | 2366 | p = real_mount(path->mnt); |
143c8c91 | 2367 | |
1da177e4 | 2368 | err = -EINVAL; |
fc7be130 | 2369 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2370 | goto out1; |
2371 | ||
5ff9d8a6 EB |
2372 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2373 | goto out1; | |
2374 | ||
1da177e4 | 2375 | err = -EINVAL; |
2d92ab3c | 2376 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2377 | goto out1; |
1da177e4 | 2378 | |
676da58d | 2379 | if (!mnt_has_parent(old)) |
21444403 | 2380 | goto out1; |
1da177e4 | 2381 | |
e36cb0b8 DH |
2382 | if (d_is_dir(path->dentry) != |
2383 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2384 | goto out1; |
2385 | /* | |
2386 | * Don't move a mount residing in a shared parent. | |
2387 | */ | |
fc7be130 | 2388 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2389 | goto out1; |
9676f0c6 RP |
2390 | /* |
2391 | * Don't move a mount tree containing unbindable mounts to a destination | |
2392 | * mount which is shared. | |
2393 | */ | |
fc7be130 | 2394 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2395 | goto out1; |
1da177e4 | 2396 | err = -ELOOP; |
fc7be130 | 2397 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2398 | if (p == old) |
21444403 | 2399 | goto out1; |
1da177e4 | 2400 | |
84d17192 | 2401 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2402 | if (err) |
21444403 | 2403 | goto out1; |
1da177e4 LT |
2404 | |
2405 | /* if the mount is moved, it should no longer be expire | |
2406 | * automatically */ | |
6776db3d | 2407 | list_del_init(&old->mnt_expire); |
1da177e4 | 2408 | out1: |
84d17192 | 2409 | unlock_mount(mp); |
1da177e4 | 2410 | out: |
1da177e4 | 2411 | if (!err) |
1a390689 | 2412 | path_put(&parent_path); |
2d92ab3c | 2413 | path_put(&old_path); |
1da177e4 LT |
2414 | return err; |
2415 | } | |
2416 | ||
9d412a43 AV |
2417 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2418 | { | |
2419 | int err; | |
2420 | const char *subtype = strchr(fstype, '.'); | |
2421 | if (subtype) { | |
2422 | subtype++; | |
2423 | err = -EINVAL; | |
2424 | if (!subtype[0]) | |
2425 | goto err; | |
2426 | } else | |
2427 | subtype = ""; | |
2428 | ||
2429 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2430 | err = -ENOMEM; | |
2431 | if (!mnt->mnt_sb->s_subtype) | |
2432 | goto err; | |
2433 | return mnt; | |
2434 | ||
2435 | err: | |
2436 | mntput(mnt); | |
2437 | return ERR_PTR(err); | |
2438 | } | |
2439 | ||
9d412a43 AV |
2440 | /* |
2441 | * add a mount into a namespace's mount tree | |
2442 | */ | |
95bc5f25 | 2443 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2444 | { |
84d17192 AV |
2445 | struct mountpoint *mp; |
2446 | struct mount *parent; | |
9d412a43 AV |
2447 | int err; |
2448 | ||
f2ebb3a9 | 2449 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2450 | |
84d17192 AV |
2451 | mp = lock_mount(path); |
2452 | if (IS_ERR(mp)) | |
2453 | return PTR_ERR(mp); | |
9d412a43 | 2454 | |
84d17192 | 2455 | parent = real_mount(path->mnt); |
9d412a43 | 2456 | err = -EINVAL; |
84d17192 | 2457 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2458 | /* that's acceptable only for automounts done in private ns */ |
2459 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2460 | goto unlock; | |
2461 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2462 | if (!parent->mnt_ns) |
156cacb1 AV |
2463 | goto unlock; |
2464 | } | |
9d412a43 AV |
2465 | |
2466 | /* Refuse the same filesystem on the same mount point */ | |
2467 | err = -EBUSY; | |
95bc5f25 | 2468 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2469 | path->mnt->mnt_root == path->dentry) |
2470 | goto unlock; | |
2471 | ||
2472 | err = -EINVAL; | |
e36cb0b8 | 2473 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2474 | goto unlock; |
2475 | ||
95bc5f25 | 2476 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2477 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2478 | |
2479 | unlock: | |
84d17192 | 2480 | unlock_mount(mp); |
9d412a43 AV |
2481 | return err; |
2482 | } | |
b1e75df4 | 2483 | |
8654df4e | 2484 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2485 | |
1da177e4 LT |
2486 | /* |
2487 | * create a new mount for userspace and request it to be added into the | |
2488 | * namespace's tree | |
2489 | */ | |
0c55cfc4 | 2490 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2491 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2492 | { |
0c55cfc4 | 2493 | struct file_system_type *type; |
1da177e4 | 2494 | struct vfsmount *mnt; |
15f9a3f3 | 2495 | int err; |
1da177e4 | 2496 | |
0c55cfc4 | 2497 | if (!fstype) |
1da177e4 LT |
2498 | return -EINVAL; |
2499 | ||
0c55cfc4 EB |
2500 | type = get_fs_type(fstype); |
2501 | if (!type) | |
2502 | return -ENODEV; | |
2503 | ||
0c55cfc4 EB |
2504 | mnt = vfs_kern_mount(type, flags, name, data); |
2505 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2506 | !mnt->mnt_sb->s_subtype) | |
2507 | mnt = fs_set_subtype(mnt, fstype); | |
2508 | ||
2509 | put_filesystem(type); | |
1da177e4 LT |
2510 | if (IS_ERR(mnt)) |
2511 | return PTR_ERR(mnt); | |
2512 | ||
8654df4e EB |
2513 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2514 | mntput(mnt); | |
2515 | return -EPERM; | |
2516 | } | |
2517 | ||
95bc5f25 | 2518 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2519 | if (err) |
2520 | mntput(mnt); | |
2521 | return err; | |
1da177e4 LT |
2522 | } |
2523 | ||
19a167af AV |
2524 | int finish_automount(struct vfsmount *m, struct path *path) |
2525 | { | |
6776db3d | 2526 | struct mount *mnt = real_mount(m); |
19a167af AV |
2527 | int err; |
2528 | /* The new mount record should have at least 2 refs to prevent it being | |
2529 | * expired before we get a chance to add it | |
2530 | */ | |
6776db3d | 2531 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2532 | |
2533 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2534 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2535 | err = -ELOOP; |
2536 | goto fail; | |
19a167af AV |
2537 | } |
2538 | ||
95bc5f25 | 2539 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2540 | if (!err) |
2541 | return 0; | |
2542 | fail: | |
2543 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2544 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2545 | namespace_lock(); |
6776db3d | 2546 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2547 | namespace_unlock(); |
19a167af | 2548 | } |
b1e75df4 AV |
2549 | mntput(m); |
2550 | mntput(m); | |
19a167af AV |
2551 | return err; |
2552 | } | |
2553 | ||
ea5b778a DH |
2554 | /** |
2555 | * mnt_set_expiry - Put a mount on an expiration list | |
2556 | * @mnt: The mount to list. | |
2557 | * @expiry_list: The list to add the mount to. | |
2558 | */ | |
2559 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2560 | { | |
97216be0 | 2561 | namespace_lock(); |
ea5b778a | 2562 | |
6776db3d | 2563 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2564 | |
97216be0 | 2565 | namespace_unlock(); |
ea5b778a DH |
2566 | } |
2567 | EXPORT_SYMBOL(mnt_set_expiry); | |
2568 | ||
1da177e4 LT |
2569 | /* |
2570 | * process a list of expirable mountpoints with the intent of discarding any | |
2571 | * mountpoints that aren't in use and haven't been touched since last we came | |
2572 | * here | |
2573 | */ | |
2574 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2575 | { | |
761d5c38 | 2576 | struct mount *mnt, *next; |
1da177e4 LT |
2577 | LIST_HEAD(graveyard); |
2578 | ||
2579 | if (list_empty(mounts)) | |
2580 | return; | |
2581 | ||
97216be0 | 2582 | namespace_lock(); |
719ea2fb | 2583 | lock_mount_hash(); |
1da177e4 LT |
2584 | |
2585 | /* extract from the expiration list every vfsmount that matches the | |
2586 | * following criteria: | |
2587 | * - only referenced by its parent vfsmount | |
2588 | * - still marked for expiry (marked on the last call here; marks are | |
2589 | * cleared by mntput()) | |
2590 | */ | |
6776db3d | 2591 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2592 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2593 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2594 | continue; |
6776db3d | 2595 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2596 | } |
bcc5c7d2 | 2597 | while (!list_empty(&graveyard)) { |
6776db3d | 2598 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2599 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2600 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2601 | } |
719ea2fb | 2602 | unlock_mount_hash(); |
3ab6abee | 2603 | namespace_unlock(); |
5528f911 TM |
2604 | } |
2605 | ||
2606 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2607 | ||
2608 | /* | |
2609 | * Ripoff of 'select_parent()' | |
2610 | * | |
2611 | * search the list of submounts for a given mountpoint, and move any | |
2612 | * shrinkable submounts to the 'graveyard' list. | |
2613 | */ | |
692afc31 | 2614 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2615 | { |
692afc31 | 2616 | struct mount *this_parent = parent; |
5528f911 TM |
2617 | struct list_head *next; |
2618 | int found = 0; | |
2619 | ||
2620 | repeat: | |
6b41d536 | 2621 | next = this_parent->mnt_mounts.next; |
5528f911 | 2622 | resume: |
6b41d536 | 2623 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2624 | struct list_head *tmp = next; |
6b41d536 | 2625 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2626 | |
2627 | next = tmp->next; | |
692afc31 | 2628 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2629 | continue; |
5528f911 TM |
2630 | /* |
2631 | * Descend a level if the d_mounts list is non-empty. | |
2632 | */ | |
6b41d536 | 2633 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2634 | this_parent = mnt; |
2635 | goto repeat; | |
2636 | } | |
1da177e4 | 2637 | |
1ab59738 | 2638 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2639 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2640 | found++; |
2641 | } | |
1da177e4 | 2642 | } |
5528f911 TM |
2643 | /* |
2644 | * All done at this level ... ascend and resume the search | |
2645 | */ | |
2646 | if (this_parent != parent) { | |
6b41d536 | 2647 | next = this_parent->mnt_child.next; |
0714a533 | 2648 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2649 | goto resume; |
2650 | } | |
2651 | return found; | |
2652 | } | |
2653 | ||
2654 | /* | |
2655 | * process a list of expirable mountpoints with the intent of discarding any | |
2656 | * submounts of a specific parent mountpoint | |
99b7db7b | 2657 | * |
48a066e7 | 2658 | * mount_lock must be held for write |
5528f911 | 2659 | */ |
b54b9be7 | 2660 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2661 | { |
2662 | LIST_HEAD(graveyard); | |
761d5c38 | 2663 | struct mount *m; |
5528f911 | 2664 | |
5528f911 | 2665 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2666 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2667 | while (!list_empty(&graveyard)) { |
761d5c38 | 2668 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2669 | mnt_expire); |
143c8c91 | 2670 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2671 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2672 | } |
2673 | } | |
1da177e4 LT |
2674 | } |
2675 | ||
1da177e4 LT |
2676 | /* |
2677 | * Some copy_from_user() implementations do not return the exact number of | |
2678 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2679 | * Note that this function differs from copy_from_user() in that it will oops | |
2680 | * on bad values of `to', rather than returning a short copy. | |
2681 | */ | |
b58fed8b RP |
2682 | static long exact_copy_from_user(void *to, const void __user * from, |
2683 | unsigned long n) | |
1da177e4 LT |
2684 | { |
2685 | char *t = to; | |
2686 | const char __user *f = from; | |
2687 | char c; | |
2688 | ||
2689 | if (!access_ok(VERIFY_READ, from, n)) | |
2690 | return n; | |
2691 | ||
2692 | while (n) { | |
2693 | if (__get_user(c, f)) { | |
2694 | memset(t, 0, n); | |
2695 | break; | |
2696 | } | |
2697 | *t++ = c; | |
2698 | f++; | |
2699 | n--; | |
2700 | } | |
2701 | return n; | |
2702 | } | |
2703 | ||
b40ef869 | 2704 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2705 | { |
2706 | int i; | |
1da177e4 | 2707 | unsigned long size; |
b40ef869 | 2708 | char *copy; |
b58fed8b | 2709 | |
1da177e4 | 2710 | if (!data) |
b40ef869 | 2711 | return NULL; |
1da177e4 | 2712 | |
b40ef869 AV |
2713 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2714 | if (!copy) | |
2715 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2716 | |
2717 | /* We only care that *some* data at the address the user | |
2718 | * gave us is valid. Just in case, we'll zero | |
2719 | * the remainder of the page. | |
2720 | */ | |
2721 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2722 | size = TASK_SIZE - (unsigned long)data; | |
2723 | if (size > PAGE_SIZE) | |
2724 | size = PAGE_SIZE; | |
2725 | ||
b40ef869 | 2726 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2727 | if (!i) { |
b40ef869 AV |
2728 | kfree(copy); |
2729 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2730 | } |
2731 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2732 | memset(copy + i, 0, PAGE_SIZE - i); |
2733 | return copy; | |
1da177e4 LT |
2734 | } |
2735 | ||
b8850d1f | 2736 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2737 | { |
b8850d1f | 2738 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2739 | } |
2740 | ||
1da177e4 LT |
2741 | /* |
2742 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2743 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2744 | * | |
2745 | * data is a (void *) that can point to any structure up to | |
2746 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2747 | * information (or be NULL). | |
2748 | * | |
2749 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2750 | * When the flags word was introduced its top half was required | |
2751 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2752 | * Therefore, if this magic number is present, it carries no information | |
2753 | * and must be discarded. | |
2754 | */ | |
5e6123f3 | 2755 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2756 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2757 | { |
2d92ab3c | 2758 | struct path path; |
1da177e4 LT |
2759 | int retval = 0; |
2760 | int mnt_flags = 0; | |
2761 | ||
2762 | /* Discard magic */ | |
2763 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2764 | flags &= ~MS_MGC_MSK; | |
2765 | ||
2766 | /* Basic sanity checks */ | |
1da177e4 LT |
2767 | if (data_page) |
2768 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2769 | ||
a27ab9f2 | 2770 | /* ... and get the mountpoint */ |
5e6123f3 | 2771 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2772 | if (retval) |
2773 | return retval; | |
2774 | ||
2775 | retval = security_sb_mount(dev_name, &path, | |
2776 | type_page, flags, data_page); | |
0d5cadb8 AV |
2777 | if (!retval && !may_mount()) |
2778 | retval = -EPERM; | |
9e8925b6 JL |
2779 | if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) |
2780 | retval = -EPERM; | |
a27ab9f2 TH |
2781 | if (retval) |
2782 | goto dput_out; | |
2783 | ||
613cbe3d AK |
2784 | /* Default to relatime unless overriden */ |
2785 | if (!(flags & MS_NOATIME)) | |
2786 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2787 | |
1da177e4 LT |
2788 | /* Separate the per-mountpoint flags */ |
2789 | if (flags & MS_NOSUID) | |
2790 | mnt_flags |= MNT_NOSUID; | |
2791 | if (flags & MS_NODEV) | |
2792 | mnt_flags |= MNT_NODEV; | |
2793 | if (flags & MS_NOEXEC) | |
2794 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2795 | if (flags & MS_NOATIME) |
2796 | mnt_flags |= MNT_NOATIME; | |
2797 | if (flags & MS_NODIRATIME) | |
2798 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2799 | if (flags & MS_STRICTATIME) |
2800 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2801 | if (flags & MS_RDONLY) |
2802 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2803 | |
ffbc6f0e EB |
2804 | /* The default atime for remount is preservation */ |
2805 | if ((flags & MS_REMOUNT) && | |
2806 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2807 | MS_STRICTATIME)) == 0)) { | |
2808 | mnt_flags &= ~MNT_ATIME_MASK; | |
2809 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2810 | } | |
2811 | ||
7a4dec53 | 2812 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 | 2813 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
c8d3e3d5 | 2814 | MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT); |
1da177e4 | 2815 | |
1da177e4 | 2816 | if (flags & MS_REMOUNT) |
2d92ab3c | 2817 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2818 | data_page); |
2819 | else if (flags & MS_BIND) | |
2d92ab3c | 2820 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2821 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2822 | retval = do_change_type(&path, flags); |
1da177e4 | 2823 | else if (flags & MS_MOVE) |
2d92ab3c | 2824 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2825 | else |
2d92ab3c | 2826 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2827 | dev_name, data_page); |
2828 | dput_out: | |
2d92ab3c | 2829 | path_put(&path); |
1da177e4 LT |
2830 | return retval; |
2831 | } | |
2832 | ||
537f7ccb EB |
2833 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2834 | { | |
2835 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2836 | } | |
2837 | ||
2838 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2839 | { | |
2840 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2841 | } | |
2842 | ||
771b1371 EB |
2843 | static void free_mnt_ns(struct mnt_namespace *ns) |
2844 | { | |
6344c433 | 2845 | ns_free_inum(&ns->ns); |
537f7ccb | 2846 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2847 | put_user_ns(ns->user_ns); |
2848 | kfree(ns); | |
2849 | } | |
2850 | ||
8823c079 EB |
2851 | /* |
2852 | * Assign a sequence number so we can detect when we attempt to bind | |
2853 | * mount a reference to an older mount namespace into the current | |
2854 | * mount namespace, preventing reference counting loops. A 64bit | |
2855 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2856 | * is effectively never, so we can ignore the possibility. | |
2857 | */ | |
2858 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2859 | ||
771b1371 | 2860 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2861 | { |
2862 | struct mnt_namespace *new_ns; | |
537f7ccb | 2863 | struct ucounts *ucounts; |
98f842e6 | 2864 | int ret; |
cf8d2c11 | 2865 | |
537f7ccb EB |
2866 | ucounts = inc_mnt_namespaces(user_ns); |
2867 | if (!ucounts) | |
df75e774 | 2868 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2869 | |
cf8d2c11 | 2870 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2871 | if (!new_ns) { |
2872 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2873 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2874 | } |
6344c433 | 2875 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2876 | if (ret) { |
2877 | kfree(new_ns); | |
537f7ccb | 2878 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2879 | return ERR_PTR(ret); |
2880 | } | |
33c42940 | 2881 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2882 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2883 | atomic_set(&new_ns->count, 1); |
2884 | new_ns->root = NULL; | |
2885 | INIT_LIST_HEAD(&new_ns->list); | |
2886 | init_waitqueue_head(&new_ns->poll); | |
2887 | new_ns->event = 0; | |
771b1371 | 2888 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2889 | new_ns->ucounts = ucounts; |
d2921684 EB |
2890 | new_ns->mounts = 0; |
2891 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2892 | return new_ns; |
2893 | } | |
2894 | ||
0766f788 | 2895 | __latent_entropy |
9559f689 AV |
2896 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2897 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2898 | { |
6b3286ed | 2899 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2900 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2901 | struct mount *p, *q; |
9559f689 | 2902 | struct mount *old; |
cb338d06 | 2903 | struct mount *new; |
7a472ef4 | 2904 | int copy_flags; |
1da177e4 | 2905 | |
9559f689 AV |
2906 | BUG_ON(!ns); |
2907 | ||
2908 | if (likely(!(flags & CLONE_NEWNS))) { | |
2909 | get_mnt_ns(ns); | |
2910 | return ns; | |
2911 | } | |
2912 | ||
2913 | old = ns->root; | |
2914 | ||
771b1371 | 2915 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2916 | if (IS_ERR(new_ns)) |
2917 | return new_ns; | |
1da177e4 | 2918 | |
97216be0 | 2919 | namespace_lock(); |
1da177e4 | 2920 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2921 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2922 | if (user_ns != ns->user_ns) |
132c94e3 | 2923 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2924 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2925 | if (IS_ERR(new)) { |
328e6d90 | 2926 | namespace_unlock(); |
771b1371 | 2927 | free_mnt_ns(new_ns); |
be34d1a3 | 2928 | return ERR_CAST(new); |
1da177e4 | 2929 | } |
be08d6d2 | 2930 | new_ns->root = new; |
1a4eeaf2 | 2931 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2932 | |
2933 | /* | |
2934 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2935 | * as belonging to new namespace. We have already acquired a private | |
2936 | * fs_struct, so tsk->fs->lock is not needed. | |
2937 | */ | |
909b0a88 | 2938 | p = old; |
cb338d06 | 2939 | q = new; |
1da177e4 | 2940 | while (p) { |
143c8c91 | 2941 | q->mnt_ns = new_ns; |
d2921684 | 2942 | new_ns->mounts++; |
9559f689 AV |
2943 | if (new_fs) { |
2944 | if (&p->mnt == new_fs->root.mnt) { | |
2945 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2946 | rootmnt = &p->mnt; |
1da177e4 | 2947 | } |
9559f689 AV |
2948 | if (&p->mnt == new_fs->pwd.mnt) { |
2949 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2950 | pwdmnt = &p->mnt; |
1da177e4 | 2951 | } |
1da177e4 | 2952 | } |
909b0a88 AV |
2953 | p = next_mnt(p, old); |
2954 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2955 | if (!q) |
2956 | break; | |
2957 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2958 | p = next_mnt(p, old); | |
1da177e4 | 2959 | } |
328e6d90 | 2960 | namespace_unlock(); |
1da177e4 | 2961 | |
1da177e4 | 2962 | if (rootmnt) |
f03c6599 | 2963 | mntput(rootmnt); |
1da177e4 | 2964 | if (pwdmnt) |
f03c6599 | 2965 | mntput(pwdmnt); |
1da177e4 | 2966 | |
741a2951 | 2967 | return new_ns; |
1da177e4 LT |
2968 | } |
2969 | ||
cf8d2c11 TM |
2970 | /** |
2971 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2972 | * @mnt: pointer to the new root filesystem mountpoint | |
2973 | */ | |
1a4eeaf2 | 2974 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2975 | { |
771b1371 | 2976 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2977 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2978 | struct mount *mnt = real_mount(m); |
2979 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2980 | new_ns->root = mnt; |
d2921684 | 2981 | new_ns->mounts++; |
b1983cd8 | 2982 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2983 | } else { |
1a4eeaf2 | 2984 | mntput(m); |
cf8d2c11 TM |
2985 | } |
2986 | return new_ns; | |
2987 | } | |
cf8d2c11 | 2988 | |
ea441d11 AV |
2989 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2990 | { | |
2991 | struct mnt_namespace *ns; | |
d31da0f0 | 2992 | struct super_block *s; |
ea441d11 AV |
2993 | struct path path; |
2994 | int err; | |
2995 | ||
2996 | ns = create_mnt_ns(mnt); | |
2997 | if (IS_ERR(ns)) | |
2998 | return ERR_CAST(ns); | |
2999 | ||
3000 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3001 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3002 | ||
3003 | put_mnt_ns(ns); | |
3004 | ||
3005 | if (err) | |
3006 | return ERR_PTR(err); | |
3007 | ||
3008 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3009 | s = path.mnt->mnt_sb; |
3010 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3011 | mntput(path.mnt); |
3012 | /* lock the sucker */ | |
d31da0f0 | 3013 | down_write(&s->s_umount); |
ea441d11 AV |
3014 | /* ... and return the root of (sub)tree on it */ |
3015 | return path.dentry; | |
3016 | } | |
3017 | EXPORT_SYMBOL(mount_subtree); | |
3018 | ||
bdc480e3 HC |
3019 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3020 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3021 | { |
eca6f534 VN |
3022 | int ret; |
3023 | char *kernel_type; | |
eca6f534 | 3024 | char *kernel_dev; |
b40ef869 | 3025 | void *options; |
1da177e4 | 3026 | |
b8850d1f TG |
3027 | kernel_type = copy_mount_string(type); |
3028 | ret = PTR_ERR(kernel_type); | |
3029 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3030 | goto out_type; |
1da177e4 | 3031 | |
b8850d1f TG |
3032 | kernel_dev = copy_mount_string(dev_name); |
3033 | ret = PTR_ERR(kernel_dev); | |
3034 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3035 | goto out_dev; |
1da177e4 | 3036 | |
b40ef869 AV |
3037 | options = copy_mount_options(data); |
3038 | ret = PTR_ERR(options); | |
3039 | if (IS_ERR(options)) | |
eca6f534 | 3040 | goto out_data; |
1da177e4 | 3041 | |
b40ef869 | 3042 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3043 | |
b40ef869 | 3044 | kfree(options); |
eca6f534 VN |
3045 | out_data: |
3046 | kfree(kernel_dev); | |
3047 | out_dev: | |
eca6f534 VN |
3048 | kfree(kernel_type); |
3049 | out_type: | |
3050 | return ret; | |
1da177e4 LT |
3051 | } |
3052 | ||
afac7cba AV |
3053 | /* |
3054 | * Return true if path is reachable from root | |
3055 | * | |
48a066e7 | 3056 | * namespace_sem or mount_lock is held |
afac7cba | 3057 | */ |
643822b4 | 3058 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3059 | const struct path *root) |
3060 | { | |
643822b4 | 3061 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3062 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3063 | mnt = mnt->mnt_parent; |
afac7cba | 3064 | } |
643822b4 | 3065 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3066 | } |
3067 | ||
640eb7e7 | 3068 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3069 | { |
25ab4c9b | 3070 | bool res; |
48a066e7 | 3071 | read_seqlock_excl(&mount_lock); |
643822b4 | 3072 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3073 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3074 | return res; |
3075 | } | |
3076 | EXPORT_SYMBOL(path_is_under); | |
3077 | ||
1da177e4 LT |
3078 | /* |
3079 | * pivot_root Semantics: | |
3080 | * Moves the root file system of the current process to the directory put_old, | |
3081 | * makes new_root as the new root file system of the current process, and sets | |
3082 | * root/cwd of all processes which had them on the current root to new_root. | |
3083 | * | |
3084 | * Restrictions: | |
3085 | * The new_root and put_old must be directories, and must not be on the | |
3086 | * same file system as the current process root. The put_old must be | |
3087 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3088 | * pointed to by put_old must yield the same directory as new_root. No other | |
3089 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3090 | * | |
4a0d11fa NB |
3091 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3092 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3093 | * in this situation. | |
3094 | * | |
1da177e4 LT |
3095 | * Notes: |
3096 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3097 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3098 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3099 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3100 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3101 | * first. | |
3102 | */ | |
3480b257 HC |
3103 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3104 | const char __user *, put_old) | |
1da177e4 | 3105 | { |
2d8f3038 | 3106 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3107 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3108 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3109 | int error; |
3110 | ||
9b40bc90 | 3111 | if (!may_mount()) |
1da177e4 LT |
3112 | return -EPERM; |
3113 | ||
2d8f3038 | 3114 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3115 | if (error) |
3116 | goto out0; | |
1da177e4 | 3117 | |
2d8f3038 | 3118 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3119 | if (error) |
3120 | goto out1; | |
3121 | ||
2d8f3038 | 3122 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3123 | if (error) |
3124 | goto out2; | |
1da177e4 | 3125 | |
f7ad3c6b | 3126 | get_fs_root(current->fs, &root); |
84d17192 AV |
3127 | old_mp = lock_mount(&old); |
3128 | error = PTR_ERR(old_mp); | |
3129 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3130 | goto out3; |
3131 | ||
1da177e4 | 3132 | error = -EINVAL; |
419148da AV |
3133 | new_mnt = real_mount(new.mnt); |
3134 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3135 | old_mnt = real_mount(old.mnt); |
3136 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3137 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3138 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3139 | goto out4; |
143c8c91 | 3140 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3141 | goto out4; |
5ff9d8a6 EB |
3142 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3143 | goto out4; | |
1da177e4 | 3144 | error = -ENOENT; |
f3da392e | 3145 | if (d_unlinked(new.dentry)) |
b12cea91 | 3146 | goto out4; |
1da177e4 | 3147 | error = -EBUSY; |
84d17192 | 3148 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3149 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3150 | error = -EINVAL; |
8c3ee42e | 3151 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3152 | goto out4; /* not a mountpoint */ |
676da58d | 3153 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3154 | goto out4; /* not attached */ |
84d17192 | 3155 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3156 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3157 | goto out4; /* not a mountpoint */ |
676da58d | 3158 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3159 | goto out4; /* not attached */ |
4ac91378 | 3160 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3161 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3162 | goto out4; |
0d082601 EB |
3163 | /* make certain new is below the root */ |
3164 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3165 | goto out4; | |
84d17192 | 3166 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3167 | lock_mount_hash(); |
419148da AV |
3168 | detach_mnt(new_mnt, &parent_path); |
3169 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3170 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3171 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3172 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3173 | } | |
4ac91378 | 3174 | /* mount old root on put_old */ |
84d17192 | 3175 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3176 | /* mount new_root on / */ |
84d17192 | 3177 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3178 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3179 | /* A moved mount should not expire automatically */ |
3180 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3181 | put_mountpoint(root_mp); |
719ea2fb | 3182 | unlock_mount_hash(); |
2d8f3038 | 3183 | chroot_fs_refs(&root, &new); |
1da177e4 | 3184 | error = 0; |
b12cea91 | 3185 | out4: |
84d17192 | 3186 | unlock_mount(old_mp); |
b12cea91 AV |
3187 | if (!error) { |
3188 | path_put(&root_parent); | |
3189 | path_put(&parent_path); | |
3190 | } | |
3191 | out3: | |
8c3ee42e | 3192 | path_put(&root); |
b12cea91 | 3193 | out2: |
2d8f3038 | 3194 | path_put(&old); |
1da177e4 | 3195 | out1: |
2d8f3038 | 3196 | path_put(&new); |
1da177e4 | 3197 | out0: |
1da177e4 | 3198 | return error; |
1da177e4 LT |
3199 | } |
3200 | ||
3201 | static void __init init_mount_tree(void) | |
3202 | { | |
3203 | struct vfsmount *mnt; | |
6b3286ed | 3204 | struct mnt_namespace *ns; |
ac748a09 | 3205 | struct path root; |
0c55cfc4 | 3206 | struct file_system_type *type; |
1da177e4 | 3207 | |
0c55cfc4 EB |
3208 | type = get_fs_type("rootfs"); |
3209 | if (!type) | |
3210 | panic("Can't find rootfs type"); | |
3211 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3212 | put_filesystem(type); | |
1da177e4 LT |
3213 | if (IS_ERR(mnt)) |
3214 | panic("Can't create rootfs"); | |
b3e19d92 | 3215 | |
3b22edc5 TM |
3216 | ns = create_mnt_ns(mnt); |
3217 | if (IS_ERR(ns)) | |
1da177e4 | 3218 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3219 | |
3220 | init_task.nsproxy->mnt_ns = ns; | |
3221 | get_mnt_ns(ns); | |
3222 | ||
be08d6d2 AV |
3223 | root.mnt = mnt; |
3224 | root.dentry = mnt->mnt_root; | |
da362b09 | 3225 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3226 | |
3227 | set_fs_pwd(current->fs, &root); | |
3228 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3229 | } |
3230 | ||
74bf17cf | 3231 | void __init mnt_init(void) |
1da177e4 | 3232 | { |
13f14b4d | 3233 | unsigned u; |
15a67dd8 | 3234 | int err; |
1da177e4 | 3235 | |
7d6fec45 | 3236 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3237 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3238 | |
0818bf27 | 3239 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3240 | sizeof(struct hlist_head), |
0818bf27 AV |
3241 | mhash_entries, 19, |
3242 | 0, | |
3243 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3244 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3245 | sizeof(struct hlist_head), | |
3246 | mphash_entries, 19, | |
3247 | 0, | |
3248 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3249 | |
84d17192 | 3250 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3251 | panic("Failed to allocate mount hash table\n"); |
3252 | ||
0818bf27 | 3253 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3254 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3255 | for (u = 0; u <= mp_hash_mask; u++) |
3256 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3257 | |
4b93dc9b TH |
3258 | kernfs_init(); |
3259 | ||
15a67dd8 RD |
3260 | err = sysfs_init(); |
3261 | if (err) | |
3262 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3263 | __func__, err); |
00d26666 GKH |
3264 | fs_kobj = kobject_create_and_add("fs", NULL); |
3265 | if (!fs_kobj) | |
8e24eea7 | 3266 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3267 | init_rootfs(); |
3268 | init_mount_tree(); | |
3269 | } | |
3270 | ||
616511d0 | 3271 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3272 | { |
d498b25a | 3273 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3274 | return; |
7b00ed6f | 3275 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3276 | free_mnt_ns(ns); |
1da177e4 | 3277 | } |
9d412a43 AV |
3278 | |
3279 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3280 | { | |
423e0ab0 TC |
3281 | struct vfsmount *mnt; |
3282 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3283 | if (!IS_ERR(mnt)) { | |
3284 | /* | |
3285 | * it is a longterm mount, don't release mnt until | |
3286 | * we unmount before file sys is unregistered | |
3287 | */ | |
f7a99c5b | 3288 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3289 | } |
3290 | return mnt; | |
9d412a43 AV |
3291 | } |
3292 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3293 | |
3294 | void kern_unmount(struct vfsmount *mnt) | |
3295 | { | |
3296 | /* release long term mount so mount point can be released */ | |
3297 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3298 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3299 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3300 | mntput(mnt); |
3301 | } | |
3302 | } | |
3303 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3304 | |
3305 | bool our_mnt(struct vfsmount *mnt) | |
3306 | { | |
143c8c91 | 3307 | return check_mnt(real_mount(mnt)); |
02125a82 | 3308 | } |
8823c079 | 3309 | |
3151527e EB |
3310 | bool current_chrooted(void) |
3311 | { | |
3312 | /* Does the current process have a non-standard root */ | |
3313 | struct path ns_root; | |
3314 | struct path fs_root; | |
3315 | bool chrooted; | |
3316 | ||
3317 | /* Find the namespace root */ | |
3318 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3319 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3320 | path_get(&ns_root); | |
3321 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3322 | ; | |
3323 | ||
3324 | get_fs_root(current->fs, &fs_root); | |
3325 | ||
3326 | chrooted = !path_equal(&fs_root, &ns_root); | |
3327 | ||
3328 | path_put(&fs_root); | |
3329 | path_put(&ns_root); | |
3330 | ||
3331 | return chrooted; | |
3332 | } | |
3333 | ||
8654df4e EB |
3334 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3335 | int *new_mnt_flags) | |
87a8ebd6 | 3336 | { |
8c6cf9cc | 3337 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3338 | struct mount *mnt; |
e51db735 | 3339 | bool visible = false; |
87a8ebd6 | 3340 | |
44bb4385 | 3341 | down_read(&namespace_sem); |
87a8ebd6 | 3342 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3343 | struct mount *child; |
77b1a97d EB |
3344 | int mnt_flags; |
3345 | ||
8654df4e | 3346 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3347 | continue; |
3348 | ||
7e96c1b0 EB |
3349 | /* This mount is not fully visible if it's root directory |
3350 | * is not the root directory of the filesystem. | |
3351 | */ | |
3352 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3353 | continue; | |
3354 | ||
a1935c17 | 3355 | /* A local view of the mount flags */ |
77b1a97d | 3356 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3357 | |
695e9df0 EB |
3358 | /* Don't miss readonly hidden in the superblock flags */ |
3359 | if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) | |
3360 | mnt_flags |= MNT_LOCK_READONLY; | |
3361 | ||
8c6cf9cc EB |
3362 | /* Verify the mount flags are equal to or more permissive |
3363 | * than the proposed new mount. | |
3364 | */ | |
77b1a97d | 3365 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3366 | !(new_flags & MNT_READONLY)) |
3367 | continue; | |
77b1a97d EB |
3368 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3369 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3370 | continue; |
3371 | ||
ceeb0e5d EB |
3372 | /* This mount is not fully visible if there are any |
3373 | * locked child mounts that cover anything except for | |
3374 | * empty directories. | |
e51db735 EB |
3375 | */ |
3376 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3377 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3378 | /* Only worry about locked mounts */ |
d71ed6c9 | 3379 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3380 | continue; |
7236c85e EB |
3381 | /* Is the directory permanetly empty? */ |
3382 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3383 | goto next; |
87a8ebd6 | 3384 | } |
8c6cf9cc | 3385 | /* Preserve the locked attributes */ |
77b1a97d | 3386 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3387 | MNT_LOCK_ATIME); |
e51db735 EB |
3388 | visible = true; |
3389 | goto found; | |
3390 | next: ; | |
87a8ebd6 | 3391 | } |
e51db735 | 3392 | found: |
44bb4385 | 3393 | up_read(&namespace_sem); |
e51db735 | 3394 | return visible; |
87a8ebd6 EB |
3395 | } |
3396 | ||
8654df4e EB |
3397 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3398 | { | |
a1935c17 | 3399 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3400 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3401 | unsigned long s_iflags; | |
3402 | ||
3403 | if (ns->user_ns == &init_user_ns) | |
3404 | return false; | |
3405 | ||
3406 | /* Can this filesystem be too revealing? */ | |
3407 | s_iflags = mnt->mnt_sb->s_iflags; | |
3408 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3409 | return false; | |
3410 | ||
a1935c17 EB |
3411 | if ((s_iflags & required_iflags) != required_iflags) { |
3412 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3413 | required_iflags); | |
3414 | return true; | |
3415 | } | |
3416 | ||
8654df4e EB |
3417 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3418 | } | |
3419 | ||
380cf5ba AL |
3420 | bool mnt_may_suid(struct vfsmount *mnt) |
3421 | { | |
3422 | /* | |
3423 | * Foreign mounts (accessed via fchdir or through /proc | |
3424 | * symlinks) are always treated as if they are nosuid. This | |
3425 | * prevents namespaces from trusting potentially unsafe | |
3426 | * suid/sgid bits, file caps, or security labels that originate | |
3427 | * in other namespaces. | |
3428 | */ | |
3429 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3430 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3431 | } | |
3432 | ||
64964528 | 3433 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3434 | { |
58be2825 | 3435 | struct ns_common *ns = NULL; |
8823c079 EB |
3436 | struct nsproxy *nsproxy; |
3437 | ||
728dba3a EB |
3438 | task_lock(task); |
3439 | nsproxy = task->nsproxy; | |
8823c079 | 3440 | if (nsproxy) { |
58be2825 AV |
3441 | ns = &nsproxy->mnt_ns->ns; |
3442 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3443 | } |
728dba3a | 3444 | task_unlock(task); |
8823c079 EB |
3445 | |
3446 | return ns; | |
3447 | } | |
3448 | ||
64964528 | 3449 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3450 | { |
58be2825 | 3451 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3452 | } |
3453 | ||
64964528 | 3454 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3455 | { |
3456 | struct fs_struct *fs = current->fs; | |
58be2825 | 3457 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3458 | struct path root; |
3459 | ||
0c55cfc4 | 3460 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3461 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3462 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3463 | return -EPERM; |
8823c079 EB |
3464 | |
3465 | if (fs->users != 1) | |
3466 | return -EINVAL; | |
3467 | ||
3468 | get_mnt_ns(mnt_ns); | |
3469 | put_mnt_ns(nsproxy->mnt_ns); | |
3470 | nsproxy->mnt_ns = mnt_ns; | |
3471 | ||
3472 | /* Find the root */ | |
3473 | root.mnt = &mnt_ns->root->mnt; | |
3474 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3475 | path_get(&root); | |
3476 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3477 | ; | |
3478 | ||
3479 | /* Update the pwd and root */ | |
3480 | set_fs_pwd(fs, &root); | |
3481 | set_fs_root(fs, &root); | |
3482 | ||
3483 | path_put(&root); | |
3484 | return 0; | |
3485 | } | |
3486 | ||
bcac25a5 AV |
3487 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3488 | { | |
3489 | return to_mnt_ns(ns)->user_ns; | |
3490 | } | |
3491 | ||
8823c079 EB |
3492 | const struct proc_ns_operations mntns_operations = { |
3493 | .name = "mnt", | |
3494 | .type = CLONE_NEWNS, | |
3495 | .get = mntns_get, | |
3496 | .put = mntns_put, | |
3497 | .install = mntns_install, | |
bcac25a5 | 3498 | .owner = mntns_owner, |
8823c079 | 3499 | }; |